Artificial Intelligence Review (2023) 56:4929–5021

https://doi.org/10.1007/s10462-022-10286-2

AI‑big data analytics for building automation

and management systems: a survey, actual challenges
and future perspectives

Yassine Himeur1,2 · Mariam Elnour1 · Fodil Fadli1 · Nader Meskin3 · Ioan Petri4 ·
Yacine Rezgui4 · Faycal Bensaali3 · Abbes Amira5,6

Published online: 15 October 2022

© The Author(s) 2022

Abstract
In theory, building automation and management systems (BAMSs) can provide all the
components and functionalities required for analyzing and operating buildings. However,
in reality, these systems can only ensure the control of heating ventilation and air condi-
tioning system systems. Therefore, many other tasks are left to the operator, e.g. evaluating
buildings’ performance, detecting abnormal energy consumption, identifying the changes
needed to improve efficiency, ensuring the security and privacy of end-users, etc. To that
end, there has been a movement for developing artificial intelligence (AI) big data analytic
tools as they offer various new and tailor-made solutions that are incredibly appropriate
for practical buildings’ management. Typically, they can help the operator in (i) analyzing
the tons of connected equipment data; and; (ii) making intelligent, efficient, and on-time
decisions to improve the buildings’ performance. This paper presents a comprehensive sys-
tematic survey on using AI-big data analytics in BAMSs. It covers various AI-based tasks,
e.g. load forecasting, water management, indoor environmental quality monitoring, occu-
pancy detection, etc. The first part of this paper adopts a well-designed taxonomy to over-
view existing frameworks. A comprehensive review is conducted about different aspects,
including the learning process, building environment, computing platforms, and applica-
tion scenario. Moving on, a critical discussion is performed to identify current challenges.
The second part aims at providing the reader with insights into the real-world application
of AI-big data analytics. Thus, three case studies that demonstrate the use of AI-big data
analytics in BAMSs are presented, focusing on energy anomaly detection in residential and
office buildings and energy and performance optimization in sports facilities. Lastly, future
directions and valuable recommendations are identified to improve the performance and
reliability of BAMSs in intelligent buildings.

Keywords Artificial intelligence · Big data analytics · Building automation and

management system · Deep learning · Evaluation metrics · Computing platforms

* Yassine Himeur
[email protected]
Extended author information available on the last page of the article

13
Vol.:(0123456789)
4930 Y. Himeur et al.

Abbreviations
A3C Asynchronous advantage actor-critic
AC Actor-critic
AE Auto-encoder
AHU Air handling unit
AMI Advanced metering infrastructures
ANN Artificial neural network
AR Auto-regressive
Adaboost Adaptive Boosting
BAAKR Bagged auto-associative kernel regression
BAMS Building automation and management system
BBN Bayesian belief networks
BBRT Bootstrap bagging of regression trees
BDT Binary decision tree
BESN Bagged echo state network
BMCDT Binary multiclass-classification decision tree
BN Bayesian networks
BNN Bagging neural network
BPNN Back propagation neural network
BRT Bagged regression tree
BSA Backtracking search algorithm
BT Bagged tree
BiLSTM Bidirectional long short-termmemory
CAGR​ Compound annual growth rate
CART​ Classification and regression tree
CCTV Closed-circuit televi-sion
CDT CART decision tree
CIFG Coupled input and forget gate
CNN Convolutional neural network
CPS Cyber-physical system
CRBM Conditional restricted Boltzmann machines
CRF Conditional random fields
CRGP Compact regression Gaussian process
CRT​ Completely-random tree
CUSP Computational urban sustainability platform
ConvLSTM Convolutional LSTM
DBN Deep belief network
DBSCAN Density-based spatial clustering of applications with noise
DCRF Deep cascade forest
DDPG Deep deterministic policy gradient
DDoS Distributed denial of service
DFNN Deep feed forward neural networks
DL Deep learning
DNN Deep neural networks
DQL Deep Q-learning
DRED Dutch resi-dential energy dataset
DRL Deep reinforcement learning
DT Decision tree
DWT Discrete wavelet transform

13
AI‑big data analytics for building automation and management… 4931

EBT Ensemble bagging tree

ELM Extreme learning machine
FCM Fuzzy C-means
FCRBM Factored conditional restricted Boltzmann machines
FFNN Feed-forward neural network
GAM Generalized additive models
GBM Gradient boosting machine
GBRT Gradient boosting regression tree
GLRM Generalized linear regression model
GPRM Gaussian process regression model
GRU​ Gated recurrent units
GSA Gravitational search algorithm
GSD Generated sampled data
HCA Hierarchical cluster analysis
HEP Hydro-electric power
HMM Hidden Markov model
HVAC Heating ventilation and air conditioning system
ICT Information and communication technology
IEQ Indoor environmental quality
Isomap Isometric feature mapping
KNN K-nearest neighbors
KPCA Kernel principal component analysis
LDA Linear discriminant analysis
LOF Local outlier factor
LR Linear regression
LR Logistic regre-ssion
LSSVR Least squares support vector regression
LSTM Long short-term memory
MC Monte Carlo
MCSVM Multi-class support vector machines
MDA Multiple discriminant analysis
MDS Multidimensional scaling
MLP Multi-layer perceptron
MLR Multiple linear regression
MPC Model predictive control
MSA Multiplicative season algorithm
MetaFA Metaheuristic firefly algorithm
NB Naive Bayes
NILM Non-intrusive load monitoring
NN-SAE Neural network-based-supervised auto-encoder
NN Neural network
OCSVM One-class support vector machine
PCA Principal component analysis
PG Policy gradient
PLS Partial least square
PPO Proximal policy optimization
QDA Quadratic discriminant analysis
QL Q-leaning
QR Quantile regression

13
4932 Y. Himeur et al.

RBDT Regression binary decision tree

RBFNN Radial basis function neural network
RBM Restricted Boltzmann machines
RF Random forest
RFT Regression fitting
RL Reinforcement learning
RNN Recurrent neural networks
RT Regression tree
ResNet Deep residual network
SAE Memory-gated RNN-based autoencoders
SAE Sparse autoencoders
SARIMA Seasonal autoregressive integrated moving average
SARSA State-action-reward-state-action
SDN Software-defined networks
SGPR Stepwise Gaussian processes regression
SSL Semi-supervised learning
SSNN Semi-supervised neural network
SVM Support vector machine
SVR Support vector regression
TRL Tradi-tional reinforcement learning
TSVD Truncated singular value decomposition
TTS Text-to-speech
VAE Variational autoencoders
VHT Vertical hoeffding tree
WQP Water quality prediction
XGBM Extreme gradient boosting machine
XGBoost EXtreme gradient boosting
mLSTM Multip-licative LSTM
t-SNE t-Distributed stochastic neighbor embedding

1 Introduction

1.1 Preliminary

Building automation and management systems (BAMSs) are intelligent systems of both
hardware and software, connecting heating ventilation and air conditioning system (HVAC)
systems, lighting, security, and other systems to communicate on a single platform. That
said, BAMSs deliver crucial information to operators and/or users on the operational per-
formance of buildings, which aim at promoting energy efficiency and optimizing water
consumption, enhancing the safety and comfort of the occupants, reducing maintenance
costs, extending the life cycle of the utilities, etc (Ippolito et al. 2014). This is possible
by networking a plethora of sensors and components responsible for the monitoring and
operation of mechanical, security, fire, lighting, HVAC and humidity control and ventila-
tion systems (Su and Wang 2020).
With the broad utilization of information and communication technologies (ICTs), sens-
ing and measurement technologies along with the cloud computing, big data storage and
data analytics, conventional BAMSs are being revolutionized. Vast quantities of building

13
AI‑big data analytics for building automation and management… 4933

automation and management data are produced, gathered and saved (Sardianos et al. 2020;
Himeur et al.). This has offered an excellent opportunity for implementing big data min-
ing and analysis in BAMSs. In this context, as the quantity of data collected in BAMSs is
enormous, the ”big data” phenomena is surfacing this field and revolutionizing the way
we manage data by using AI-big data analytics tools (Quinn et al. 2020, Himeur et al.).
Accordingly, with advanced sensing and metering technologies in BAMSs, data split into
multiple modalities and many variables can create a comprehensive source of informa-
tion to analyze. This allows for more targeted analysis, but also means that more powerful,
intelligent, and sophisticated tools are needed to identify the most enormous patterns/varia-
bles (Muntean et al. 2021). As a consequence, the big data analytics market in the building
energy sector is expected to grow at a Compound annual growth rate (CAGR) of 11.28%,
during the forecast period, 2021–2026.1 Data collection in the building industry is becom-
ing all-embracing. This wealth of big data allows informed data-driven decision-making by
designers, facilities managers, and owners during building design, operation, and retrofit
(Berger et al.). On the other hand, for existing or outdated buildings to make full use of
the services offered by the flourishing data analytics market, the necessary enhancement
to the existing system for deploying the new technology must be addressed and sorted out
(Varlamis et al. 2022). The main challenges of gathering and analyzing data of old build-
ings are the outdated technologies and the conventional error-prone data collection means
(Jia et al. 2019; Al Dakheel et al. 2020). Nevertheless, data analytics can assist in designing
and implementing the new system adaptation and existing system renovation (Elnour et al.
2022).
Besides, it is of utmost importance to know the current state of AI-based building
automation before presenting the actual study concerning user input, demand, response,
energy-saving, and automation. In this respect, it is obvious that AI adds new dimensions
to building automation environments by enabling autonomous data analysis for operation
optimization. Therefore, many AI-based contributions have recently emerged as key solu-
tions for (i) predicting building occupancy, (ii) forecasting thermal comfort, (iii) boosting
energy saving, and (iv) enabling demand-side response (Himeur et al. 2020). Additionally,
as mentioned in previous studies, (O’Grady et al. 2021) people can spend up to 90 percent
of their lives in buildings; this highlights the importance of user input, behavioral data,
and behavioral analytics for optimizing and automating building operations. To that end,
a significant research effort is ongoing to develop AI-based behavioral change technolo-
gies to promote energy saving in residential and office buildings (Sayed et al., Varlamis
et al. 2022), understand consumers demand patterns for successful demand response devel-
opment (Cruz et al. 2021; Pratt and Erickson 2020), optimizing occupants’ thermal com-
fort (Zheng et al. 2022), transforming water management (Doorn 2021), improving fault
detection and diagnosis (Yun et al. 2021), etc. Moreover, AI-based big data analytics are
contributing to building automation by making BAMSs self-learning, self-configuring and
self-diagnosing, and self-commissioning (Katipamula 2019). Additionally, using AI-based
analytics can adapt existing building systems to promote the deployment of BAMSs with
fewer investments from building owners.
From another hand, as AI models are very competent to learn common human error
patterns, their use in big data analytics is significant. They can (i) detect and resolve pos-
sible flaws in datasets, (ii) learn by watching how the operators and users interact with

1
https://​www.​resea​rchan​dmark​ets.​com/​repor​ts/​47749​56/​big-​data-​analy​tics-​market-​in-​the-​energy-​sector.

13
4934 Y. Himeur et al.

the analytics programs, and identify anomalies and surface unexpected insights from large-
scale datasets fast (Mahmud et al. 2020; Diamantoulakis et al. 2015). In this context, AI
models assist operators and users of BAMSs to perform the different tasks related to the
big data cycle, among them the operations of collecting, pre-processing, aggregating, stor-
ing, analyzing and extracting various kinds of features (Hu and Vasilakos 2016; Bode et al.
2019). Moving on, the integration of AI-big data analytics can (i) optimize energy and
operational efficiency, (ii) automate monitoring and control through wireless platforms,
(iii) provide quick and better decision making, (iv) smartly control the facility and reduce
risk failures, (v) lower life cycle costs, and (vi) increase safety and security measures with
ease (Aghemo et al. 2014; Zhou and Yang 2016; Aste et al. 2017).

1.2 Paper contributions

Due to the importance of using AI-big data analytics in BAMSs, a plethora of works have
been proposed to (i) address different challenges, (ii) improve and automate building oper-
ation, and (iii) optimize building user experience. In addition, different reviews have been
introduced to discuss the advances made in this research topic, such as (Zhang et al. 2021;
Molina-Solana et al. 2017; Zhao et al. 2020). However, most of them have only focused
on addressing one task at a time, e.g., energy management, rather than covering multiple
BAMS tasks together (e.g., water management, occupancy detection, comfort optimiza-
tion, fault diagnosis and anomaly detection (FDAD), etc.) (Sun et al. 2020; Wang et al.
2021; Fan et al. 2018). To that end, we present in this paper a comprehensive systematic
survey reflecting the latest developments in the field of AI-big data analytics and their uti-
lization in BAMSs from different perspectives. For example, Zhang et al. (2021) discuss
sensor impact verification and evaluation for FDAD in energy systems, while Molina et al.
(2017) review the contributions of data science for building energy management issues.
Moving on, data mining strategies used for building energy management are overviewed
in Zhao et al. (2020). Similarly, in Sun et al. (2020), data-driven techniques for energy
prediction in buildings are described. Besides, Wang et al. (2021) focus on studying the
practical problems related to implementing ML models for building energy efficiency. It
also investigates the commitment of existing studies to comfort and energy saving (i.e.,
Save energy with/without compromising thermal comfort). Moreover, in Fan et al. (2018),
unsupervised data mining methodologies for energy efficiency improvement are analyzed.
Lastly, in Pinto et al. (2022), Pinto et al. discuss the roles of transfer learning integration
for smart buildings and systems.
To that end, we present in this paper a comprehensive survey reflecting the latest devel-
opments in the field of AI-big data analytics and their utilization in BAMSs from different
perspectives. Thus, we first introduce a generic taxonomy for classifying AI-big data ana-
lytics frameworks based on various criteria, including the learning method, building envi-
ronment, computing platform, and application or challenge addressed. Typically, an over-
view of existing works and discussions is presented, highlighting some of the challenges,
limitations, and shortcomings. Then, three case studies are presented illustrating the use of
AI-big data analytics for critical concerns in the buildings sector, that is, energy efficiency
and management, to provide the reader with insight into real-world applications. The opti-
mization of energy consumption in buildings has been a hot research topic recently2 in

2
https://​www.​energy.​gov/​sites/​prod/​files/​2017/​03/​f34/​qtr-​2015-​chapt​er5.​pdf.

13
AI‑big data analytics for building automation and management… 4935

terms of efficient planning, proper management, and preventive maintenance. Lastly, future
directions to ease the use of AI-big data analytics models in BAMSs and improve their
feedback are derived. To summarize, the contributions of the presented work are manifold:

– Providing a thorough review covering the general use of AI-big big data analytics in
BAMSs and shedding light on their increasing importance for developing efficient and
smart BAMSs.
– Presenting a well-designed taxonomy of existing AI-big data analytics frameworks,
which helps in understanding intriguing relationships between various concepts and
variables in the field. Different criteria have been adopted when analyzing existing
frameworks, including the learning method, building environment, computing platform,
application, etc.
– Conducting a critical analysis and discussion to (i) extract diverse relevant lessons that
are learned from overviewed works; and (ii) highlight open issues and current chal-
lenges, among them data scarcity, data benchmarking, security and privacy, scalability
and interoperability and real-time big data intelligence.
– Presenting three case studies that describe the use of AI-big data analytics in BAMSs
for buildings energy management and optimization, such that the first two case studies
demonstrate unsupervised and supervised energy anomaly detection strategies in resi-
dential and office buildings, and the third one is about energy and performance optimi-
zation in sports facilities.
– Deriving a set of future research and development directions that attract considerable
interest in the near and far future, and help in improving the performance and reliability
of BAMSs.

Table 1 outlines some of the main differences between the actual review and other survey
studies. It also sheds light on some of the main contributions addressed by this review
compared to the others in terms of overviewed resources (i.e., ML tools and computing
platforms), application scenarios, discussed challenges (i.e., security issues), evaluation
metrics, case studies, and proposed future directions (i.e., multimodal data analysis, in-situ
sensor calibration in BAMSs, smart building digital twins, blockchain edge analytics, etc.).

1.3 Review methodology

A well-established review methodology is adopted in this paper, where we first con-

duct a comprehensive literature search in the most popular scientific databases, includ-
ing Scopus, Elsevier, Wiley, and IEEE. Following, most of the works that deal with the
use of AI-big data analytics for BAMSs are included in this study. Many keywords and
their combination are then used in the search, e.g., ”building automation and manage-
ment systems”, ”big data analytics”, ”artificial intelligence”, ”machine learning”, ”deep
learning”, ”transfer learning”, ”energy prediction in buildings using machine learning”,
”thermal comfort in building using machine learning”, ”fault diagnosis and anomaly
detection in buildings”, ”security in building automation and management systems”,
etc. Therefore, research studies introduced between January 2015 and February 2022

13
 4936

Table 1  Comparison of the proposed survey’s contributions against other existing related related review studies

13
ML tools Com- Application sce- Security issues Evaluation Case Future directions
puting narios metrics studies
platforms Multimodal Blockchain Edgeanalytics 3D
data analysis point
clouds

Zhang et al. ✗ ✗ FDAD ✗ ✗ ✗ ✗ ✗ ✗ ✗

(2021)
Molina-Solana ✓ ✓ Energy manage- ✓ ✗ ✗ ✗ ✗ ✗ ✗
et al. (2017) ment
Zhao et al. (2020) ✓ ✓ Energy manage- ✗ ✗ ✗ ✗ ✗ ✗ ✗
ment
Sun et al. (2020) ✓ ✗ Energy prediction ✗ ✗ ✗ ✗ ✗ ✗ ✗
Wang et al. (2021) ✓ ✗ Energy efficiency ✗ ✗ ✗ ✓ ✗ ✗ ✗
Fan et al. (2018) ✓ (unsupervised) ✗ Energy efficiency ✗ ✗ ✗ ✗ ✗ ✗ ✗
Pinto et al. (2022) ✓ (transfer learn- ✗ Load prediction, ✗ ✓ ✗ ✗ ✗ ✗ ✗
ing) system control,
occupancy detec-
tion, building
dynamics
Our paper ✓ ✓ Energy manage- ✓ ✓ ✓ ✓ ✓ ✓ ✓
ment FDAD,
IEQ, security and
safety, occupancy
detection and
water manage-
ment
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4937

Fig. 1  The significant progress made in the development of BAMSs during the last two decades

are discussed in this framework. This period has arbitrarily been selected to evaluate the
recent and pertinent contributions. Typically, this framework discusses English-written
peer-reviewed journal articles, conference proceedings papers, and book chapters. The
selection process adopted in this review relies on adhering to the specifications of the
PRISMA (Moher et al. 2009), which is a practical and efficient approach for writing
survey studies. Concretely, a search was performed for the last seven years (January
2015–February 2022). To eliminate duplicate references, a reference manager software
was utilized, and only the remaining frameworks have then been considered after filter-
ing them by their titles, keywords, and abstracts.
In addition to reviewing existing AI-big data analytics contributions for BAMSs, three
case studies are also included in this article to provide the reader with more explanations
about using AI tools in tackling the buildings’ energy consumption question in terms of (i)
unsupervised energy anomaly detection, (ii) supervised energy anomaly detection, and (iii)
energy and performance optimization for sports facilities.

1.4 Organization of the paper

The rest of this paper is structured as follows. Section 2 highlights the significant advances
made in the development of BAMSs. Section 3 provides an overview of the interdiscipli-
nary AI-big data analytics research in BAMSs following a well-defined taxonomy. Sec-
tion 4 evaluates and critically analyses overviewed frameworks to identify the open issues
and current challenges. Section 5 presents three case studies that describe the use of AI-big
data analytics in BAMSs for energy anomaly detection in residential and office buildings,
and energy optimization in sports facilities. Moving on, Sect. 6 presents the future direc-
tions for improving the performance of BAMSs. Finally, conclusions and significant find-
ings are summarized in Sect. 7.

2 Evolution of BAMSs

In the last decades, BAMSs have been rapidly developed; indeed, from the 1950s to 1990s,
they have been transformed from pneumatics to electronics then open protocols (e.g. BAC-
nets). Moving forward, with the digitization era, BAMSs have further progressed by (i)
integrating more powerful and smart technologies, (ii) becoming easier to implement in
different kinds of buildings, and (iii) using high-quality softwares to aid the users in getting
the most pertinent information from their buildings. Indeed, digitization has significantly

13
4938 Y. Himeur et al.

accelerated with the launch of smartphones, where these devices have abruptly replaced
mobile communicators, cell phones, etc. and become more practical in many real-world
applications as they deliver different benefits, such as supporting apps.

2.1 Progress made during the two last decades

We briefly described in this section the most significant achievements made in the devel-
opment of BAMSs during the last two decades, which can be summarized as portrayed
in Fig. 1. Typically, in 2008, it became possible to virtualize BAMSs in data centers,
and hence receives greater security and availability and enables more flexible access to
buildings’ data. In 2009, WiFi was integrated to BAMSs to help in flexibly and remotely
monitoring appliances in households and commercial centers (Wang 2009). Following,
in 2010, due to the growing utilization of smartphones and tablet computers in smart
city applications to control smart and location-based products, BAMSs have been also
positively influenced through developing more sophisticated and portable BAMSs solu-
tions (Aste et al. 2017).
By 2014, audio files had been deployed in BAMSs using the text-to-speech (TTS)
technology. The latter enabled to support preventive inspections and maintenance, work
contracts, service requests, work contracts, equipment audits, etc (Mundt et al. 2014). In
2016, IoT began to significantly influence the society after that IoT devices found their
way into the building management sector, where six billion devices were installed and
more than 31 billion were expected in 2020 (Aste et al. 2017). Explicitly, the impor-
tance of automation has been increased in both existing and new buildings. Follow-
ing, in 2018, commercial buildings have progressively evolved into smart buildings
and routines have been improved and/or automated in BAMSs, resulting in enhanced
comfort and efficiency (Markoska and Lazarova-Molnar 2018; Lee and Karava 2020).
Lastly, in 2020, AI joined the spectrum of BAMSs to help in early fire detection, and
energy demand prediction. Moreover, it becomes possible to identify behavioral change
through the analysis of real-time data. Thus, BAMSs learn from experiences and histori-
cal data for automatically adjusting the indoor conditions (Yaïci et al. 2021).

2.2 Big data sources

This section discusses and describes the essential sources of heterogeneous big data
used to implement AI-big data analytics in BAMSs. Indeed, developing and accelerat-
ing the advance and deployment of BAMSs require the installation of a large number
of smart sensors, smart meters, and other measurement devices in the different parts
of each building, which helps in (i) increasing the observability of its transient and
dynamic events, and (ii) gather actual data related to the diverse functionalities of the
building. This will later help the AI-big data analytics in accurately analyzing this data
and extracting pertinent features and therefore facilitating the operation and monitoring
of all building technology, especially in larger buildings. Figure 2 portrays the overall
architecture of a BAMS and its principal data sources. The control module is the central
brain of the BAMS, and most of the controllers are built using the industry standard
BACnet protocols in addition to Konnex (KNX); an open communication standard for
commercial and domestic building automation, LonWorks; a standardized bus system
used in centralized and decentralized building automation control (Merz et al. 2018),

13
AI‑big data analytics for building automation and management… 4939

Fig. 2  Principal services of a BAMS system

and Modbus; a network communication protocol for connecting electronic equipment in

industrial automation systems. Overall, a BAMS can provide various services to control
(i) heating and cooling, (ii) lighting, (iii) security, (iv) access control, (v) fire and life
safety, and (vi) elevators and escalators.

3 Overview of AI‑big data analytic frameworks

3.1 Overall taxonomy

To understand the challenges related to AI-big data analytics in BAMSs, it is essen-

tial to perform a generic taxonomy of existing AI-big data analytics techniques used
for monitoring the smart buildings. Specifically, Fig. 3 provides a structured analysis
framework that helps in overviewing existing techniques and shedding the light on the
organization of the presented framework.

3.2 AI‑learning process

The first step in any AI process is system learning. This can take four primary forms:
supervised learning, unsupervised learning, semi-supervised, and reinforcement learn-
ing. In this section, we present an overview of existing AI learning architectures used to
improve the performance of BAMSs.

13
4940 Y. Himeur et al.

Fig. 3  Taxonomy of existing AI-big data analytics frameworks

3.2.1 Unsupervised learning (U)

Unsupervised learning learns from raw data without prior knowledge and mainly deals
with unlabeled datasets. Although it does not need to annotate data as supervised learn-
ing, the learning phase can be more computational as all the possibilities are checked.
The accuracy is lower since there are no corresponding outputs (labels) (Himeur et al.
2021).

3.2.1.1 U1. Clustering It is a category of ML algorithms used for separating data (e.g.
energy consumption observations, ambient conditions, etc.) into different classes or clus-
ters following a specific goal. Clustering algorithms usually pertain to one of the follow-
ing groups, i.e. hybrid, fuzzy-based, model-based, and density-based approaches. Using the
clustering process facilitates the classification tasks when dealing with various problems,
such as anomaly detection of energy consumption, indoor environmental quality (IEQ)
monitoring and detection of pollutants, detection of abnormal water consumption, etc.
K-means, C-means and fuzzy C-means (FCM) were among the most investigated clus-
tering approaches. They have been applied for non-intrusive load monitoring (NILM) and
appliance identification (Ji et al. 2019; Zhang et al. 2020), energy performance evalua-
tion and ranking in working spaces (Sun and Yu 2021), energy efficiency assessment in

13
AI‑big data analytics for building automation and management… 4941

industrial buildings (Liu et al. 2018), building management and identification of operat-
ing anomalies (when analyzing electricity, gas and water consumption) (Akil et al. 2019),
IEQ monitoring (Dogruparmak et al. 2014; Cao et al. 2020; Alghamdi et al. 2020; Roger
Rozario 2021), energy forecasting (Tian et al. 2020; Chen et al. 2020; El Motaki et al.
2021), and data sampling for better visualization (Qin and Zhang 2017).
In Culaba et al. (2020), an energy prediction model is introduced using a k-means model
to cluster data, and a support vector machine (SVM) is employed to forecast energy con-
sumption. In Himeur et al. (2021), three clustering algorithms, namely one-class support
vector machine (OCSVM), density-based spatial clustering of applications with noise
(DBSCAN), and local outlier factor (LOF), are used to detect anomalous energy consump-
tion in households by analyzing energy footprints. Besides, in Afaifia et al. (2021), hierar-
chical cluster analysis (HCA) is implemented to model residential energy consumption and
promote energy efficiency. Clustering-based techniques have been used in BAMSs because
of their simplicity and relatively computational efficiency. Also, clustering models gener-
ally have few parameters to tune. However, they have different limitations that affect their
applications in BAMSs, among them the manual selection of the optimal K, dependency
on initial values, troubles to cluster data with varying densities and sizes, the need for scal-
ing as the number of dimensions increases, etc. (Li et al. 2018).

3.2.1.2 U2. Dimensionality reduction In diverse ML tasks, dimensionality reduction tech-

niques can be employed to classify data while promoting low computational costs as they
first remove irrelevant observations. Accordingly, a plethora of frameworks have been pro-
posed in the literature to explore the applicability of dimensionality reduction schemes in
BAMSs. That includes the principal component analysis (PCA), factor analysis, linear dis-
criminant analysis (LDA), quadratic discriminant analysis (QDA), and multiple discrimi-
nant analysis (MDA), isometric feature mapping (Isomap) (Liu et al. 2020), kernel principal
component analysis (KPCA) (Abba et al. 2020), t-distributed stochastic neighbor embed-
ding (t-SNE) (Zhan et al. 2020; Lopes et al. 2020), multidimensional scaling (MDS) (Wang
2020, and truncated singular value decomposition (TSVD) (Kalantzis et al. 2021).
For instance, PCA has been utilized for early fault detection and classification (Li and
Wen 2014; Cotrufo and Zmeureanu 2016; Chen and Wen 2017; Swiercz and Mroczkowska
2019), IEQ (Mansor et al. 2021), energy consumption prediction (Sha et al. 2019), occu-
pancy detection in buildings (Pal et al. 2019), etc. Moving on, LDA has been employed
for thermal comfort evaluation (Gładyszewska-Fiedoruk and Sulewska 2020), sensor-based
occupancy detection (Fayed et al. 2019). Using dimensionality reduction in BAMS has
gained attention because it can: (i) reduce the storage space and time needed to classify
recorded data, (ii) improve the interpretation of the ML models’ parameters by removing
multicollinearity, and (iii) simplify data visualization (Al-Kababji et al. 2022). However,
dimensionality reduction models have some disadvantages. For example, (i) this can result
in relevant data loss, (ii) finding linear correlations between variables (as PCA does) can be
no appropriate in some scenarios, and (iii) some dimensionality reduction models can fail
in classifying variables if the covariance and mean are not sufficient to represent datasets
(Himeur et al. 2021; Abdulhammed et al. 2019).

13
4942 Y. Himeur et al.

3.2.2 Supervised learning (S)

Supervised learning is applied for the case of labeled energy datasets. Despite its high per-
formance, the necessity of labeled data causes some difficulties in real-world applications.

3.2.2.1 S1. Classification It refers to conventional ML models that attempt to derive some
conclusions from the input data given in the training process, and hence aim at predicting
the class labels/categories for a new set of data. Classification models have widely been
deployed in existing BAMS based big data analytics frameworks to perform different tasks,
e.g. energy forecasting, energy balancing, IAQ monitoring, energy optimization, fault and
anomaly detection. Typically, SVM, K-nearest neighbors (KNN) (Valgaev et al. 2017), deci-
sion tree (DT) (Yu et al. 2010), artificial neural network (ANN) (Moon et al. 2019), multi-
layer perceptron (MLP) (Haidar et al. 2019), extreme learning machine (ELM) (Salerno and
Rabbeni 2018) and logistic regression (LR) (Rehman et al. 2020) are among the famous
classification models deployed in BAMSs. Classification models have been used in BAMSs
since they are simple to understand, fast and efficient. In addition, they can excel in clas-
sifying different kinds of BAMS data if accurately labeled datasets are used in the train-
ing process. However, they have a set of limitations. For instance, SVM models are not
adequate for non-linear problems, and their performance does not improve if the number of
features increases while the number of neighbors ”K” is manually selected in KNN. Moreo-
ver, poor results are usually obtained with DT algorithms on small datasets, and overfitting
can quickly occur (Himeur et al. 2021).

3.2.2.2 S2. Regression It is based on the identification of the relation between two or more
energy consumption observations for producing a set of model parameters, that help in pre-
dicting and classifying them for different purposes, including energy prediction, anomaly
detection, security and privacy preservation, etc. Diverse regression models have been pro-
posed to analyze BAMSs’ data, e.g. support vector regression (SVR) (Zhong et al. 2019),
linear regression (LR), auto-regressive (AR) models, regression tree (RT) and regression fit-
ting (RFT). Regression models have gained popularity in smart buildings and smart energy
systems because most of them are easy to implement and interpret, and efficient to train.
Also, they perform remarkably well for linearly separable data. However, it is worthy to note
that regression models involve complicated and lengthy procedures of analysis and calcula-
tions in addition to assuming the existence of linearity between the dependent and independ-
ent variables, which is not always the case in real-world applications (Bilous et al. 2018).

3.2.2.3 S3. Deep neural networks (DNN) It is a subclass of ML that is principally a NN

including more than two layers. These DNNs aim at simulating the behavior of the human
brain ”albeit far from matching its ability”, which allows them to learn from large-scale
datasets. In addition to the capability of NNs with a single layer for making approximate
predictions, DNNs have further benefits via (i) optimizing and refining the classification
accuracy when additional hidden layers are considered, and (ii) identifying the most inform-
ative features of the data.
DNNs have become the state-of-the-art methods in various ML-based domains, simi-
larly in BAMSs, they are attracting greater attention. They have widely used for energy
forecasting, this is the case of recurrent neural networks (RNN), long short-term memory
(LSTM) (Gao et al. 2019; Wang et al. 2020), gated recurrent unit (GRU) (Lin et al. 2021),
bidirectional LSTM (BiLSTM) (Haq et al. 2021; Ishaq and Kwon 2021), convolutional

13
AI‑big data analytics for building automation and management… 4943

LSTM (ConvLSTM) (Syed et al. 2021), multiplicative LSTM (mLSTM) (Krause et al.),
bidirectional GRU (BiGRU) (Khan et al. 2020), coupled input and forget gate (CIFG)
(Runge and Zmeureanu), deep feed forward neural networks (DFNN) (Marino et al. 2016),
and convolutional neural network (CNN) (Li et al. 2017). Moreover, numerous hybrid
models have been built by combining the aforementioned models with other deep learning
(DL) architectures, such as CNN-LSTM (Alhussein et al. 2020), CNN-BiLSTM (Wu et al.
2021), partial least square (PLS) CNN-BiLSTM (PLS-CNN-BiLSTM) (Wu et al. 2021),
CNN-GRU (Sajjad et al. 2020; Wu et al. 2020), conditional random fields (CRF) and RNN
(CRF-RNN) (Wytock and Kolter 2013), DFNN-LSTM (Bashari and Rahimi-Kian 2020),
radial basis function neural network-CNN (RBFNN-CNN) (Sideratos et al. 2020), etc.
DL models have also been used for other tasks, including smart IEQ monitoring, where
different architectures were investigated, such as LSTM (Liu et al. 2020; Janarthanan et al.
2021), GRU (Ahn et al. 2017; Das et al. 2020), BiLSTM (Ma et al. 2019), CNN (Moli-
nara et al. 2020), residual neural network (ResNet) (Zhang et al. 2020), variational autoen-
coders (VAE) coupled with CNN (VAE-CNN) (Loy-Benitez et al. 2020a), memory-gated
RNN-based autoencoders (MG-RNN-AE) (Loy-Benitez et al. 2020b), sparse autoencoders
(SAE) (Loy-Benitez et al. 2020)
Occupancy detection in buildings has also received the attention of the DL commu-
nity through the use of CNN (Zou et al. 2017), RNN (Zhao et al. 2018), LSTM (Mutis
et al. 2020), BiLSTM (Feng et al. 2020). Moreover, as some studies have investigated
the use of camera imagery (e.g. thermal cameras) to estimate the number of occupants
inside buildings, it was rational to use various CNN backbones, which are widely utilized
in image classification or image recognition, among them ResNet (Acquaah et al. 2020),
VGGNet (Zou et al. 2017), AlexNet (Acquaah et al. 2020) and GoogLeNet (Tien et al.
2020). Using DL models in BAMSs has become a research hot-spot nowadays because of
their robustness to natural variations in the data, which is automatically learned. Addition-
ally, their performance significantly improves with increasing the quantity of training data.
However, DL models still face different challenges. Typically, DL algorithms require large-
scale training datasets to perform better than other ML models. Moreover, their training is
computationally expensive as they are built on complex models. Additionally, DL models
require expensive GPUs and cloud data centers to run, which increases their deployment
cost (Guo et al. 2018; Himeur et al. 2020).

3.2.2.4 S4. Statistical models They refer to mathematical models embodying an ensem-
ble of statistical rules used to generate data samples, predict the relationships between one
or diverse random/non-random variables or classify them. widely used statistical models
include Bayesian networks (BN) (Singh and Yassine 2018), naive Bayes (NB) (Li et al.
2020), generalized additive models (GAM) (Khamma et al. 2020), bayesian belief networks
(BBN) (Bassamzadeh and Ghanem 2017), restricted Boltzmann machines (Elsaeidy et al.
2019), conditional restricted Boltzmann machines (CRBM) (Kang et al. 2020) and factored
conditional restricted Boltzmann machines (FCRBM) (Hafeez et al. 2020). In BAMSs,
they have been used for different tasks, such as selecting the most energy-efficient primary
HVAC systems (Tian et al. 2019), building energy and water retrofitting (Bertone et al.
2018), energy forecasting (Huang et al. 2018), assessing energy efficiency (Grillone et al.
2019), NILM (Verma et al. 2019), gas usage prediction (Pathak et al. 2018), IEQ monitor-
ing (Giovanis 2019), etc. While most statistical models are useful for BAMSs as they have
deterministic and stochastic components to mathematically describe the functional relation-
ship between inputs and outputs, they also have pitfalls. In this respect, if recorded data is

13
4944 Y. Himeur et al.

biased or faulty, statistical modeling will be misleading. In addition, these kinds of models
are hard to apply to heterogeneous data (Agha and Palmskog 2018).

3.2.3 Semi‑supervised learning (SSL)

SSL refers to the process of training ML models using a small portion of labeled data along
with a large number of unlabeled observations. Then, the ML models should be able to
learn and make predictions on new data. It falls between unsupervised learning and super-
vised learning, which is also considered as a special instance of weak supervision (Van
Engelen and Hoos 2020). Although supervised learning techniques are largely utilized in
for providing different BMAS services, they can only reach high performance only when
they are trained with sufficient labeled data. Otherwise, their performance could drastically
decrease if annotated data is insufficient or not accurately labeled. Moreover, annotating
data is a challenging, costly, and time-consuming task. In this regard, SSL has been pro-
posed as an alternative solution to address some of these issues.
In BAMS, SSL has been widely used for fault and anomaly detection. For instance, in
Fan et al. (2021, 2021), the authors introduce an SSL-based fault detection and diagnosis
in air handling units (AHUs) based on a semi-supervised neural network (SSNN), which
adopts a self-training strategy. Moving on, in Elnour et al. (2021), Elnour et al. propose an
SSL-based data-driven attack detection scheme in HVAC systems to promote security in
intelligent buildings. This approach has been developed using an isolation forest and two
ML models, i.e. PCA and 1D-CNN (IF-PCA-CNN). While in Li et al. (2021), an SSL-
based approach to detect and diagnose chiller faults is presented using a semi generative
adversarial network (semi-GAN) model. In the same way, an SSL-based fault identifica-
tion scheme for building HVAC systems is proposed in Li et al. (2021) using a modified
GAN. In Nguyen et al. (2021), an SSL-based load monitoring solution is introduced, which
has the ability to (i) augment the data, (ii) transform existing labeled sets, and (iii) train a
WiderResNet (the backbone model) on the augmented data. Although SSL is an excellent
option for developing AI-big data analytics when labeled data is expensive to obtain, it has
some limitations, e.g., the results are not stable, and the performance is lower than that of
supervised learning. Typically, the decision boundary might be overstrained if the training
dataset does not have the annotated samples required in each class (Lu 2009).

3.2.4 Reinforcement learning (RL)

Reinforcement learning is a field in artificial intelligence that involves an agent that devel-
ops the knowledge of the best strategy to follow to accomplish a defined objective by trial
and error given the interaction with its environment. Besides the RL agent and the envi-
ronment, the main elements of an RL system are: (i) the policy, which is a function that
defines the action taken by the RL agent in a given time step (i.e., state), (ii) the reward,
which defines the result of the action taken by the agent due to its interaction with the
environment, and intuitively describes the desired behavior of the agent, and (iii) the value,
which indicates the long-term desirability of a set of states/actions given the agent’s expe-
rience and the likely future rewards (Collins and Cockburn 2020). The agent explores the
possible actions to be taken as the learning progresses. Based on the consequence of the
actions taken, it opts for actions that maximize the cumulative reward. Reinforcement
learning algorithms can be categorized as: (i) traditional RL (TRL) methods in which

13
AI‑big data analytics for building automation and management… 4945

tabular (i.e. lookup tables) or conventional value function approximation approaches (e.g.,
coarse coding, ML algorithms) are used; and (ii) deep RL (DRL), which represents the
evolution of the traditional methods where DL models (e.g., deep NNs, CNNs, RNNs) are
used to approximate the state and/or action value (Wang and Hong 2020).

3.2.4.1 R1. TRL models It is only efficient to use TRL for simple RL problems where the
action-state space can be represented in a tabular form or approximated by a simple func-
tion approximation algorithm. Monte Carlo (MC), Q-leaning (QL), State-action-reward-
state-action (SARSA), policy gradient (PG), and actor-critic (AC) are examples of TRL
approaches. For TRL-based BAMS applications, tabular QL was used for occupancy pre-
diction and HVAC control to optimize the occupant comfort and energy consumption in
Barrett and Linder (2015), and controlling the HVAC system and windows for mechanical
and natural ventilation in Chen et al. (2018).

3.2.4.2 R2. DRL models Recently, RL has taken advantage of the DL technology to reach
phenomenal results. Typically, DL has been combined with RL due to its ability to capture
all the intricate details of the knowledge and also perform complicated learning tasks that
RL failed in doing so. This has given rise to DRL. In BAMSs and many other research
fields, DRL is becoming a significant focus of scientists. The commonly used DRL methods
are deep Q-learning (DQL), asynchronous advantage actor-critic (A3C), deep deterministic
policy gradient (DDPG), and proximal policy optimization (PPO). A review of DRL appli-
cations for intelligent buildings energy management was presented in Yu et al. DQL was
used for indoor and domestic hot water temperature control in Lissa et al. (2021) to optimize
the home energy management system. In Wang et al. (2017), a DRL-based control system
for office HVAC systems using an RNN-based actor-critic approach was presented. In Val-
ladares et al. (2019), double Q-learning was utilized for energy optimization and thermal
comfort control, while PPO method was applied in Azuatalam et al. (2020); Chemingui
et al. (2020) for controlling the building’s HVAC systems for energy and thermal comfort
optimization.
AL in all, RL models (TRL and DRL) are utilized for solving very complex problems
that can not be fixed using traditional ML or DL models. They can also correct the errors
occurring during the training stage. However, exceeding the number of required RL stages
can result in an overload of states, and hence reducing the performance of RL models
(Ding et al. 2019).

3.2.5 Ensemble methods (E)

Ensemble methods are a class of ML that deploy different aggregation strategies for com-
bining multiple learning models and then achieving better predictive performance com-
pared to the use of a unique learning algorithm.

3.2.5.1 E1. Boosting It implies the gradual development of an ensemble learning using
a set of ML models, where every new model occurrence is trained for emphasizing the
training occurrences that previous models misclassified. In some applications, boosting can
achieve better performance than bagging; however, it often looks after overfitting the train-
ing data. Random forest (RF), adaptive Boosting (Adaboost), and eXtreme gradient boost-
ing (XGBoost) were among the most used boosting models for different AI-big data analyt-

13
4946 Y. Himeur et al.

ics tasks, such as overall building energy consumption forecasting (Zekić-Sušac et al. 2021;
Xiao et al. 2021; Ferdoush et al.; Wang and Chen 2021; Yucong and Bo 2020), heating and
ventilation load prediction (Sun et al. 2020), HVAC optimization (Li 2020), Space cooling
load forecasting (Feng et al. 2021), load disaggregation and monitoring (Xiao et al. 2021),
water monitoring (Somontina et al. 2018; Movahedi and Derrible; Golabi et al. 2020), IEQ
monitoring (Mo et al. 2019).
Following, other variants have been then introduced and utilized for performing energy
forecasting and load monitoring, IEQ monitoring, water management and occupancy
detection in different kinds of buildings, including gradient boosting machine (GBM)
(Gong et al. 2020), extreme gradient boosting machine (XGBM) (Gong et al. 2020), gradi-
ent boosting regression tree (GBRT) (Nie et al. 2021), LightGBM (Park et al. 2021; Wang
et al. 2020).

3.2.5.2 E2. Bootstrap aggregating It is also abbreviated as bagging and refers to the design
of a new ML model by aggregating multiple models that have equal weights in the ensem-
ble vote. Every model is trained using a randomly drawn subclass of training data for pro-
moting the model’s variance. Various bagging models have been developed, modified and
used to perform different tasks in BAMSs. For instance, bagging ARIMA (BARIMA) in de
Oliveira and Oliveira (2018) is proposed to conduct a mid-long term load forecasting, while
in Khwaja et al. (2015), a bagging neural network (BNN) is developed where the bagging
concept is combined with neural networks (NNs) to improve short-term energy prediction.
Moving on, in Hu et al. (2020), an enhanced bagged echo state network (BESN) is intro-
duced to forecast energy. In Choi and Hur (2020), a bagging model is developed by setting
RF, XGBoost and LightGBMs as the base learners. In Dehalwar et al. (2016), the authors
introduce a bagged regression tree (BRT) that has been used for energy forecasting.
Moreover, bagging models have also been employed for water management in buildings
using ensemble bagging tree (EBT) (Hasanzadeh Nafari et al. 2016), and thermal evalu-
ation using bagged tree (BT) (Ahmad and Chen 2018), and fault detection using bagged
auto-associative kernel regression (BAAKR) (Yu et al. 2017).
Overall, ensemble methods have been used in BAMSs since they can result in better
predictive accuracy than individual models in complex systems/models Moreover, they are
appropriate for scenarios with linear and non-linear data variables. However, ensembling is
less interpretable, and the outputs of ensemble models are complex to explain and predict
in most applications. In addition, a wrong selection of the models to be aggregated will
arise lower predictive accuracy than individual models. Furthermore, ensemble models are
generally computationally expensive and require much storage memory.

3.3 Building environments and their characteristics

Buildings range in size, function, construction, design, and other attributes. Additionally,
they present varying levels of potential hazards and risks to the occupants and the sur-
rounding environment. However, buildings are primarily classified based on the utilization
purpose that governs occupancy profile, sophistication level, and building design require-
ments. Building environments are further described in the following subsections.

13
AI‑big data analytics for building automation and management… 4947

3.3.1 Residential buildings

Residential buildings are mainly for private occupancy, designed and built for individu-
als or groups, providing the necessary facilities and utilities to satisfy living requirements.
Spaces in residential buildings involve several activities, including sleeping, sitting, con-
veniences, cooking, dining, and others. Those functions can be in shared spaces or have
exclusive rooms per function. They exist in various sizes and have different occupancy
rates. A low occupancy density generally characterizes them. Examples of residential
buildings are story houses, apartments, terraces, and condominiums. In addition to air con-
ditioning and ventilation systems, lighting, and media equipment, several major appliances
are regularly used in residential buildings, such as dishwashers, washers, dryers, refrigera-
tors, freezers, stoves, water heaters, trash compactors, ovens, and others (Estiri 2014).
Residential buildings are typically equipped with simple BAMSs that provide the basic
requirements of building management for inhabitants’ well-being and comfort. Standard
manual control is used for the most part of their BAMSs. For instance, the decentralized
control of the indoor environment is driven by the thermal comfort levels of the occupants.
Thermal comfort is subjective to outside weather conditions that determine the indoor
environment conditioning requirements, which are heating or cooling, humidification or
dehumidification, and air ventilation (Do and Cetin 2018).

3.3.2 Office buildings

Office buildings are where people perform routine tasks, execute assignments and jobs
for their employers, or provide passive or active, free of charge or remunerated ser-
vices to the public. Types of workplaces vary in the form and requirements of the work
and the variety of tools involved. Hence, they differ in size and the extent of personnel
involvement and expertise. Familiar workplaces are office buildings such as law and
corporate firms, commercial companies, post offices, banks, courtrooms, and similar
places where people are involved in lengthy desk jobs or light-weight activities. Most
of the spaces are offices, meeting rooms, or auditoriums of defined capacities. Addi-
tionally, they have shared areas such as corridors and lobbies. Office buildings require
flexible and technologically-advanced working environments that are safe, healthy,
pleasant, durable, and accessible towards promoting the users’ comfort, productivity,
and working efficiency (Tanabe et al. 2013). It includes the accessibility to natural
ventilation and natural lighting sources, and the availability of IEQ control and moni-
toring. The provision of localized indoor environment control allows users to adjust
the air temperature, air movement, and other relevant indoor environment properties
according to their preferences. They are characterized by their moderate operation
schedules, and fairly regular and established user profiles. Additionally, some work-
spaces may involve many service recipients (Alsalemi et al.).

3.3.3 Healthcare centers

The indoor environment in healthcare centers is critical for the health, well-being,
safety, and comfort of patients, visitors, and the staff, as well as for the medical utili-
ties and services. It has to comply with specific standards related to temperature, infec-
tion, and odor control (Salonen et al. 2013). It plays a significant role in the quality of

13
4948 Y. Himeur et al.

the provided medical service in terms of the treatment, healing, recovery processes,
and the success of the conducted operations and procedures.
The various spaces in healthcare centers have different temperature regulation
requirements. For instance, the success of surgical procedures depends in part upon
the cold indoor conditions of the operating room to avoid the risks of anesthetic explo-
sions, promote the comfort, productivity, and efficiency of the staff, and conserve the
patient’s resources (Ellis 1963). On the other hand, burn units are regulated at temper-
atures between 28 and 33 degrees because burn injuries restrict the ability of patients’
bodies to stay warm (Fernández and Pablo 2021). Moreover, healthcare centers require
a clean and sterile environment. Hospital-acquired infections are a major threat in
healthcare centers (Lobdell et al. 2012). Hence, air ventilation and infection control
are essential to control the potential contaminants and other suspended microorgan-
isms, consequently lowering airborne disease risk. Additionally, air ventilation helps
dispel odors, which improves the indoor conditions for the patients, staff, and visitors.
Moreover, medical waste disposal and management is an essential aspect of the opera-
tion of healthcare centers as they are considered one of the main sites for the genera-
tion of hazardous waste (Aljabre 2002). The proper management of medical waste is
essential to avoid health and environmental risks. Healthcare centers have protocols for
the disposal of the generated waste according to their location. Additionally, health-
care centers are obliged to provide adequate security implementations for (i) the safety
of patients, the public, and staff, (ii) the privacy and integrity of the patients’ data, (iii)
the prevention of breaches against the BAMS, (iv) the management of the utilities and
equipment, and (v) the prevention of injuries and unwanted occurrences.

3.3.4 Sports facilities

Sports facilities involve areas where individuals or groups engage in physical exercise,
participate in athletic competitions, or attend sporting events. Examples of sports facilities
are gymnasiums, cultural centers, stadiums, swimming pools, indoor and outdoor tennis
courts, squash courts, training halls, and sports arenas. They encompass large and various
spaces involving different types of activities. Sports facilities have distinct requirements for
air conditioning and ventilation, thermal comfort, and lighting with unique usage and occu-
pancy patterns. They are governed by the type of sports activity, the operating time, the
season, and the geographical location of the facility (Trianti-Stourna et al. 1998).
Sports facilities are characterized by the variety of their architectural sophistication
and sizes, deployed technologies, and their distinctive energy demand profile compared to
other types of buildings (Elnour et al. 2022). For instance, stadiums are the most sophis-
ticated ones, which occupy vast land space. Even though they are often infrequently used,
their operation and running costs during a single event are substantial (Aquino and Nawari
2015). Aquatic centers are the second most popular sports facilities that host different water
events and tournaments. They encompass other spaces such as changing rooms, shower
rooms, and storage rooms.
Sustainability measures and implementation are deployed in sports facilities’ design,
construction, and operations. They require extensive lighting, air conditioning, broadcast-
ing, surveillance, and security requirements when operated to achieve successful sports
events. The proper lighting in the sports facilities ensures good visual conditions. The
event’s prosperous broadcasting is essential to delivering an entertaining, thrilling, and
engaging experience for the athletes and fans. Given the considerable volume of user flow

13
AI‑big data analytics for building automation and management… 4949

in sports facilities, emergency evacuation planning, users’ entry and exit management,
security screening, and preventive measures are among the top priorities in sports facili-
ties management (Hall et al. 2011). Additionally, sports facilities involve extensive body
workouts and activity by the users, during which excessive heat and CO2 discharge occur.
They demand mainly air conditioning and ventilation, especially for indoor sports events
as well as water heating for pools and domestic use to maintain the comfort, health, and
well-being of the users. Moreover, sports facilities require constant maintenance, servicing,
and overseeing even when not used, such as grass fields, pools, water treatment, and sports
equipment.

3.3.5 Commercial buildings

Commercial buildings have at least 50% of their floor spaces for commercial activities
(Kiliccote and Piette), such as malls, retail, and food services. Malls and restaurants are
typical commercial buildings of various sizes and complexity. They include shops, cafes,
kitchens with several commercial appliances, storage rooms, pantries, a refrigerated space,
offices, dining areas, and public restrooms. They demand maintaining a clean and well-
conditioned environment. For example, in restaurants and coffee shops, compliance with
proper food storage and preparation standards is required to reduce the risk of spoiling food
and eliminate the risk of incidents jeopardizing the well-being of the users as well as the
reputation of the restaurants (El-Sharkawy and Javed 2018).
Air ventilation affects the health and safety of workers and customers and can influence
food sanitation levels. The chiefs and the kitchen staff in restaurants are exposed to air pol-
lutants generated from cooking for long periods. Hence, they may suffer potential respira-
tory and cardiovascular problems in the long run (Juntarawijit and Juntarawijit 2017). Also,
they are subjected to high levels of heat generated from cooking activities, decreasing the
staff’s productivity. Additionally, excessive unpleasant odors or poor air conditioning in
restaurants can result in an unpleasant experience for the customers. In addition, malls and
shopping centers are commercial buildings where goods or services are sold to customers.
They may include ample parking spaces, escalators, elevators, and various outlets such as
department stores, food courts, amusement and theme parks, and movie theaters. Safe and
comfortable indoor conditions are essential to provide a convenient and enjoyable experi-
ence for users and maintain a flourishing business with efficient energy consumption to
contain the incurred running and operating costs.
Commercial buildings are famous for their exceptional operating schedule and occu-
pancy patterns. They run for about more than 12 hours all week, and they have peak occu-
pancy during weekends and significant volumes of user flow. They utilize extensive closed-
circuit television (CCTV) surveillance, lighting, and air ventilation and conditioning
systems. Additionally, fire prevention, suspension, and other security and alarm systems
are crucial elements of their management systems to ensure dependable and safe circum-
stances for the users. Overly, the proper management of commercial buildings is essential
to maintain a lucrative operation.

3.3.6 Industrial buildings

Industrial buildings include buildings used for the generation and distribution of power,
manufacturing products such as food, apparel, electronics, petrochemicals, construc-
tion materials, automobiles, the processing of raw materials, and many others. They have

13
4950 Y. Himeur et al.

minimal and relatively low user flow for security purposes, such that they are only acces-
sible to individuals with privileges. However, they involve energy-intensive and delicate
machinery. They are generally equipped with sophisticated BAMSs that support the secu-
rity and the centralized control requirements. Industrial buildings are equipped with robot-
ics, industrial devices, and software-defined production processes. They require a high
level of automation, given the nature of the processes involved and the tasks performed.
In addition, they may involve delicate processes that are associated with health, social, and
environmental risks. Industrial sites and environments can result in air and water pollution
due to the generated by-products and the released unwanted toxins of the occurring pro-
cesses. Hazards from combustion and unstable reactions can lead to highly harmful acci-
dents due to the sudden release of material at high temperatures or pressures (Englund
2007). Additionally, fire hazards are common in industrial facilities, which can endanger
the lives of staff and can result in substantial economic losses and environmental implica-
tions. Industrial facilities must be safe, secure, and productive. Proper process control, air
ventilation, treatment and conditioning, and waste management are crucial to managing the
safety and health of the staff as well as the general public and the surrounding environment.
Security is an essential dimension in the operation of the BAMSs of industrial buildings.

3.3.7 Academic buildings

Academic buildings are used to conduct teaching activities such as schools, academies,
universities, colleges, technical institutes, etc. They encompass classrooms, lecture halls,
libraries, student centers, dining halls, laboratories, computer labs, offices, and service
areas necessary for the proper functioning of the academic programs. Individuals of vari-
ous age groups are frequent users of educational facilities, and they engage in multiple
types of activities. A convenient and safe environment in academic facilities is an essen-
tial requirement for the education process. It affects the well-being and comfort of stu-
dents, faculty members, and other staff, hence their productivity and working efficiency. A
comfortable and safe environment has been identified as an essential element for enhanc-
ing the learning of students (Muhammad et al. 2014). Over-heated and poorly ventilated
classrooms can result in the discomfort of students and educators, and consequently diverts
their attention and affects their abilities to concentrate (Roelofsen). Adequate lighting in
the facilities of academic buildings is vital to the comfort and well-being of the students to
create an attractive and engaging learning environment and avoid eye strain. Additionally,
students’ health and well-being are essential for their learning process. The indoor environ-
ment influences students’ attendance and hence their study. Students need to be in good
health to be able to study well. Therefore, spaces in academic buildings should be well
conditioned and ventilated to avoid altering users’ well-being, spreading airborne diseases
spread, and disrupting students’ learning. Lastly, a brief summary is presented in Table 2 to
compare the characteristics of the different buildings discussed above.

3.4 Computing platforms

3.4.1 Cloud computing

The advancement of cloud computing platforms has opened new opportunities for BAMSs
to take control of operations on a large scale. Thus, BAMSs that consist of networked

13
 Table 2  A comparison between the different types of buildings
Building type BAMS type Use Spaces type Activity type Size Occu- Characteristics Requirements Example
pancy/
Usage

Residential Simple for basic requirements Dwelling Spaces for sleep- - Dwell- Small to Low to - Light Iinhabitants’ Houses, apart-
for inhabitants’ well-being ing, sitting, ing, Light moderate mod- operations, and well-being and mentscondomini-
and comfort, standard conveniences, weight erate causal equip- comfort ums
manual control cooking, dining activities ment/appli-
ances
- Consume rela-
tively substan-
tial amounts
of energy and
accompa-
nied by peak
demand issues
Office Standard for natural ventila- Business Offices, meet- - Routine Small to Moder- - Fairly regular healthy, flexible, Firms, banks,post
tion, natural lighting, ing rooms, tasks, moderate ate to and established durable, produc- offices
local indoor environment auditoriums, services to high user profiles tive, efficient,
control, surveillance and corridors, public and comfort-
AI‑big data analytics for building automation and management…

safety system lobbies - Desk jobs able working

or light- environment
weight
activities
4951

13
 Table 2  (continued)
4952

Building type BAMS type Use Spaces type Activity type Size Occu- Characteristics Requirements Example
pancy/

13
Usage

Heathcare Advanced for air treatment, Health Treatment rooms, - Delicate Aver- Moder- - Year-round -Health, well- Hospitals, clinic-
ventilation, and temperature services care units, medical age to ate to operation, fre- being, safety, shealth centers
regulation, security, surveil- examination procedures large high quent and con- and comfort of
lance, servicesequipment rooms, labo- - Lab experi- siderable users patients, visitors,
management ratories, blood ments flow- include the staff, medi-
banks, operat- sophisticated cal utilities and
ing rooms, and expensive services-comply
patients rooms, equipment with specific
waiting rooms, for diagnostic standards related
corridors, and treatment- to temperature,
lobbies associated infection, and
with health and odor control.
environmental - Clean and sterile
risks. environment
Sports Advanced system Sporting Training rooms, - Sporting Aver- Moder- -High seasonal -Distinct require- Stadiums, sports
for extensive light- events and spectators activities age to ate to usage patterns ments for air centers,swimming
ing, air conditioning, activities areas, chang- of different large high of high users conditioning pools
broadcasting,surveillance, ing rooms, levels and flow, and ventilation,
and security shower rooms, types -Encompass thermal comfort
storage rooms, various space - Specific visual
offices, lobbies, types, conditions and
corridors - Substantial run- broadcasting
ning and opera- requirements
tion costs
Y. Himeur et al.
 Table 2  (continued)
Building type BAMS type Use Spaces type Activity type Size Occu- Characteristics Requirements Example
pancy/
Usage

Commercial Inclusive system for closed- Commerce Shops, cafes, din- - Standard, Aver- Moder- Year-round -Clean and well- Shopping malls,
circuit television, surveil- ing areas, lightweight age to ate to operation conditioned grocery
lance, lighting, air ventila- kitchens, stor- activities large high with frequent environment- stores,restaurants
tion and conditioning age rooms, peak periods Proper air
offices, lobbies, of considerable ventilation for
corridors users flow maintaining
health and safety
of workers and
customers
Industrial Sophisticated system for Industries Offices, control - Light to Aver- Low to Involve limited - High level of Factories, power
centralized automation and rooms, com- heavy age to mod- and consist- automation stations
management puter rooms, weight large erate ent users flow, - safe, secure,
machinery and activities energy-inten- and productive
process rooms, sive and deli- operation
electrical and cate machinery - waste manage-
chemical plants and processes ment
AI‑big data analytics for building automation and management…

associated with
health, social,
and environ-
mental risks
Academic Inclusive system for Education Classrooms, - Light to Aver- Moder- - Seasonal usage - A convenient, Universities,
surveillance, lighting, air offices, meet- heavy age to ate to patterns of comfortable and colleges,schools
ventilation and conditioning, ing rooms, weight large high moderate to safe environment
and security dining halls, activities high flow for students and
laboratories, staff
computer - Adequate light-
labs, lobbies, ing
corridors
4953

13
4954 Y. Himeur et al.

sensors and actuators, have been recently adapted to be able to connect to different cloud-
based services (Alsalemi et al. 2020). The latter can provide data storage, connectivity, and
powerful computing resources. To that end, significant efforts have been devoted to devel-
oping cloud-based big data analytics solutions in BAMSs (Bode et al. 2019). For instance,
a voice-activated system for remotely monitoring BAMSs using cloud computing is pre-
sented in Valenzuela et al. (2013). While in Khattak et al. (2019), the idea of developing
vehicular clouds for smart buildings and smart city applications is investigated. Moving
on, in Stergiou et al. (2018), the security and privacy concerns along with the efficiency of
cloud platforms are analyzed. In Delsing (2017), local cloud IoT automation is studied to
promote the use of distributed IoT automation solutions.
Despite the significant effort made during the last decade to promote the use of cloud-
services to run BAMSs, some drawbacks are still causing issues to users and operators,
among them (i) the increased cost and communication overheads, (ii) the privacy and secu-
rity concerns, especially when private data is transmitted to a centralized server for pro-
cessing (Mohamed et al. 2018).

3.4.2 Edge computing

Edge computing it refers to performing data pre-processing, data fusion for different
sources and AI-big data analytics at the edge of the network i.e. sensor nodes (Ray et al.
2019). Also, it enables optimizing cloud computing platforms due to its capability to use
the processing power of IoT devices for filtering, pre-processing, aggregating and storing
IoT sensor data. These tasks can correspondingly be conducted in real-time using conveni-
ent analytical tools (Sharma et al. 2018), while cloud platforms perform further enrich-
ment, aggregation and running complex analytics on the filtered data. To that end, the new
advances in BAMSs combined with the latest generation of IoT devices make it possible
to bring the intelligence and computing tasks to the edge nodes in close proximity to the
building’s IoT devices (Zakharchenko and Stepanets 2019; Khan et al. 2020). Moreover, a
new generation of open software platforms hosted on edge nodes are enabling access to the
building data and advanced AI-big analytics deployed on these platforms are providing the
technology to create value from this data by transforming data from building environments
into actionable information. Various open edge platforms have recently been proposed, e.g.
IOTech’s Edge Xpert,3 Echelon SmartServer IoT platform,4 JENEsys Edge,5 etc.

3.4.3 Fog computing

Fog computing represents a decentralized computing strategy where data storage, data
processing and computing resources are located in the middle layer situated between edge
devices and cloud. Typically, IoT smart sensors and submeters periodically collect the data
and forward it to a gateway that acts as a fog device (Javadzadeh and Rahmani 2020). In
this line, BAMSs can benefit from streaming data over a layer of fog devices (or nodes) to
become more connected, where data can be analyzed to detect abnormalities for example,
and autonomously react, if authorized, for compensating the problems or fixing the issues.
Otherwise, fog nodes will send the convenient requests to the cloud (or services higher up

3
https://​www.​iotec​hsys.​com/​marke​ts/​indus​tries/​build​ing-​autom​ation/.
4
https://​www.​dialog-​semic​onduc​tor.​com/​produ​cts/​indus​trial-​edge-​compu​ting/​smart​server-​iot-​edge-​server.
5
https://​www.​lynxs​pring.​com/​techn​ology.

13
AI‑big data analytics for building automation and management… 4955

the fog hierarchy) for making further skilled and powerful technical analysis using com-
plex ML models (Ferrández-Pastor et al. 2018; Aazam et al. 2018).
For instance, in some situations that require real-time decision-making, e.g. shut down
appliances or equipment before being damaged or adjust crucial process parameters, edge
devices or fog nodes can rapidly act with millisecond-level latency, while it is not possi-
ble to reach real-time decision making using cloud data centers (Rocha Filho et al. 2018).
Therefore, the use of fog computing or edge computing helps avoid potential latency prob-
lems, delays an/or network/server down-times that can lead to different kinds of accidents
or reduced service optimization and efficiency (Maatoug et al. 2019).

3.4.4 Hybrid computing

Hybrid computing refers to the case when the aforementioned computing architectures, i.e.
edge computing, fog computing and cloud computing, are used together to process and
analyze data (Himeur et al. 2021; Zhang et al. 2020). In this context, based on the appli-
cation scenario and computation requirement, some data processing tasks could be made
at the edge devices and/or fog nodes, while high-level data processing tasks (e.g. feature
extraction, classification, anomaly detection, etc.) could be performed at the cloud data
centers (Himeur et al. 2020).

3.5 Applications

3.5.1 Facility and asset management

Facility management to eliminate waste is among the benefits of using AI-big data analyt-
ics in BAMSs and can perform in diverse forms. For instance, using an AI strategy, a bath-
room supplies monitoring company has saved up to 40% in of the total cost by installing
a sensors that collect and send information about the utilization levels of toilet paper rolls
and soaps (Gaboalapswe 2019; Sayed et al. 2022). Similar techniques are also be deployed
for monitoring sports facilities, commercial buildings, office supplies, and other building
necessities (Himeur et al. 2021; Idowu et al. 2016).
For instance, in sports facilities there is an emergency to improve the BAMS services to
meet consumer’s growing experience needs, and hence, overcome various issues, e.g. poor
resource sharing, weak flexibility of response and slow transmission of information, and
instability of aero-thermal comfort, which are considerable affecting the end-users’ expe-
rience and restricting the development of sport venues (Zhong et al. 2020). To that end,
a great attention has been put recently to design intelligent BAMS architectures of sport
centers. This helps in interconnecting multiple subsystems, improving the interoperability,
integrating information, realizing the integration of data application network, and achieves
the goal of resource sharing and function upgrading. In Xiao-wei (2020), an AI-big data
analytics platform is built using SVM-back propagation neural network (SVM-BPNN)
for (i) predicting the end-user flow in the sport facility, (ii) providing recommendations
to adjust the service plan, and (iii) improving the overall management and the end-users’

13
4956 Y. Himeur et al.

experience. Moving on, in Wan et al. (2021), as the cyber-security is a challenging issue
in sports facilities due to the number of spectators and players and the large number of
sport events organized, an AI-assisted cyber-physical system (AI-CPS) is integrated to the
BAMS for promoting network security and predicting cyber attacks and adversaries.
On the other hand, because developing an appropriate setpoint temperature for the
HVAC system is a crucial challenge, the authors in Aparicio-Ruiz et al. (2021) identify
such temperature using a KNN-based dynamic adaptive comfort technique. It relies on the
idea that occupants’ thermal comfort in a building has different acceptability levels, which
can be used for learning the comfort temperature corresponding to the average running
temperature. Thus, this helps define the adequate range of indoor temperature. While in
Carreira et al. (2018), Carreira et al. introduce a framework for tracking building end-users’
group preferences, learning from them, and automatically managing HVAC systems. This
framework is built by tracking building users using an RFID card, interacting with them on
a mobile app, computing setpoints, and sending instructions to the HVAC sub-system over
a gateway. Additionally, a K-means algorithm has been used for configuring the setpoint, in
line with a prediction based on the current building status.

3.5.2 Load forecasting

In BAMSs, forecasting energy consumption is of significant importance to enable an effec-

tive management of energy, in which AI-big data analytics techniques play an essential
role. In doing so, load patterns (and ambient conditions) are constantly collected from
diverse building smart-meters and then fed into the AI models to predict energy usage.
Because of the real-time characteristic of short-term forecasting, it has been more chal-
lenging than generic forecasting. Thus, various AI-big data analytics models have been
proposed (Chou and Tran 2018; Ahmad and Chen 2018; Seyedzadeh et al. 2018; Fathi
et al. 2020). In Pham et al. (2020), a random forests (RF) model is introduced to perform a
short-term energy load prediction at an hourly sampling rate in various buildings by using
different energy consumption datasets. In Seyedzadeh et al. (2019), the authors investigate
the performance of diverse popular ML algorithms to predict buildings heating and cooling
energy usage. Accordingly, specific tuning has been carried out for every ML algorithm
using two building energy consumption datasets generated in EnergyPlus and Ecotect. In
Ribeiro et al. (2018), a transfer learning based load prediction scheme is introduced, where
energy consumption data of different buildings are used to forecast the load of a new build-
ing. This approach can work with various ML algorithms with pre- and post-processing
phases.
In Moon et al. (2018), Moon et al. propose an energy prediction model using diverse
ML models, including ANN, SVR, and PCA-factor analysis (PCA-FA). Data from four
buildings in an academic institution have been used for evaluating the performance of
these models. In Ahmad et al. (2020), an intelligent load prediction scheme is proposed
using generated sampled data-based Gaussian process regression model (GSD-GPRM),
regression binary decision tree (RBDT), bootstrap bagging of regression trees (BBRT) and
binary multiclass classification decision tree (BMCDT). In Idowu et al. (2016), supervised
ML algorithms are used to develop a load forecasting model using SVM, regression tree,
feed-forward neural network (FFNN), and multiple linear regression (MLR). Moving on,
in Ahmad et al. (2018), diverse supervised ML models are implemented to predict energy
consumption at short, medium, and long-term levels in different building environments,

13
AI‑big data analytics for building automation and management… 4957

namely compact regression Gaussian process (CRGP), binary decision tree (BDT), gen-
eralized linear regression model (GLRM) and stepwise Gaussian processes regression
(SGPR). In Chou and Ngo (2016), a short-term based energy prediction system is proposed
using a seasonal autoregressive integrated moving average (SARIMA) model along with
a metaheuristic firefly algorithm-based least squares support vector regression (MetaFA-
LSSVR) model. Typically, this framework uses (i) the SARIMA architecture for linearing
energy observations, and (ii) the MetaFA-LSSVR model for capturing nonlinear energy
patterns.
In Li et al. (2021), a transfer-learning-based ANN scheme is developed to predict
short-term energy consumption in information-poor buildings. The efficiency of trans-
fer learning in improving the prediction accuracy has been demonstrated using limited
training data. Moving on, in Grolinger et al. (2016), a short-term energy consumption
prediction of sports facilities, which is considered as a challenging scenario due to the
variations caused by by the hosted events, is performed using NN and SVR. In Zheng
et al. (2017), a short-term energy prediction approach using an empirical mode decom-
position (EMD)-LSTM-based RNN is proposed with a Xgboost model to select feature
patterns based on a feature importance evaluation. In a similar manner, in Haq et al.
(2021), a sequential learning-based load forecasting algorithm is developed and used
in both residential and commercial buildings. Accordingly, this framework implements
a convLSTM integrated with BiLSTM (ConvLSTM-BiLSTM) and compares its perfor-
mance with various sequential models, including ConvLSTM integrated with BiLSTM,
LSTM, auto-encoder (AE), multi-layer Bi-LSTM (MBiLSTM), BiLSTM-AE, GRU,
and CNN with multilayer bidirectional GRU (CNN-MB-GRU). Fig. 4 illustrates a
flowchart of an energy forecasting system based on AI-big data analytics. Specifically,
a short-term energy consumption prediction is performed using EMD-LSTM neural
networks with a Xgboost algorithm to extract importance features. Besides, Moradza-
deh et al. (2020) propose a heating and cooling load forecasting scheme that is based
on MLP and SVR in residential buildings. These models help identify a linear map-
ping between inputs and outputs. MLP has outperformed SVR in terms of the recall
metric, where a recall of 99.93% has been achieved. To summarize, Table 3 compares
various AI-Big data analytics frameworks used for energy forecasting, in terms of AI
model, forecast horizon, building environment, year of appearance, method description
and evaluation metrics.

3.5.3 Energy efficiency

One area that can get immensely benefited from AI-big data analytics is energy effi-
ciency in buildings. This is because of the way a building consumes energy can be
quite variable, and is related to various parameters, e.g. the nature of the building,
the energy provider, the sources of the energy, the number of devices, the number of
end-users/occupants in any building, the behavior of end-users/occupants, etc. Himeur
et al. (2021); Fatema et al. (2020). Moreover, comprehending the energy consumption
habits of a building is the first and most critical step to achieve energy efficiency. This
way, AI-big data analytics and energy efficiency can go hand in hand towards the goal
of optimizing energy consumption and reducing the amount of wasted energy without
compromising the comfort level of end-users and the level of efficiency and productiv-
ity in a company or industry Sardianos et al. (2021).

13
4958 Y. Himeur et al.

Fig. 4  Flowchart of an energy forecasting system based on AI-big data analytics (Zheng et al. 2017)

In (Yu and Chiller), Yu et al. propose an open IoT cloud-based ML system, namely
AI Chiller, to promote energy efficiency in buildings by optimizing the consumption of
the HVAC system. An AI-big data analytics scheme based on an RNN-LSTM architec-
ture and to analyzing and fusing BAMS environmental footprints has been developed
and combined with a genetic algorithm to achieve 10% savings. In Al-Ali et al. (2017),
energy saving in residential buildings in the Gulf region is achieved using IoT, off-the-
shelf business intelligence, and big data analytics platforms.

3.5.4 Predictive control and thermal comfort

One solution to save buildings’ energy is using model predictive control (MPC). It aims at
developing predictive models for (i) simulating input-output interactions; and (ii) helping
users to identify optimum control actions that drive the predicted outputs to the desired
references. In this context, ML models predict energy demand and simulate MPC con-
trol techniques to save energy and optimize end-users’ comfort. Typically, these models
can provide decision bases for selecting optimal MPC control actions Serale et al. (2018);
Mariano-Hernández et al. (2021). In Gao et al. (2020), building thermal comfort control
is conducted using RL, where a deep feed-forward neural network (FNN)-based method
is introduced. The latter helps predict consumers’ thermal comfort before introducing a
deep deterministic policy gradients (DDPGs)-based scheme to optimize thermal comfort.
In Yang et al. (2020), an MPC approach based on an RNN with nonlinear autoregressive
exogenous (NARX) architecture, namely NARX-RNN, is proposed to optimize air con-
ditioning and mechanical ventilation (ACMV) in a hospital office and hence save energy
and optimize thermal comfort. Similarly, in Yang et al. (2021), the same methodology is
experimentally implemented to control the ACMV systems in office and lecture theatre
(LT) testbeds in real-time. In Chen et al. (2020), Chen et al. developed an MPC approach
based on deep transfer learning to optimize the HVAC operation in smart buildings.
Moving forward, in Bünning et al. (2020), an RF-based data predictive control (DPC)
scheme is proposed using convex optimization and affine functions. This help in controlling

13
 Table 3  A summary of AI-Big data analytics frameworks proposed for energy forecasting in buildings
Ref. AI model Forecasthorizon Building nature Year Description Evaluation metrics
RMSE MAE MAPE Others

Skomski et al. (2020) seq2seq Short-term Office 2020 Demonstrate the ✓ nRMSE
efficiency of seq2seq
RNNs for load
prediction using a
restricted feature set
Bessani et al. (2020) Bayesian networks Short-term Residential 2020 Handle the volatility ✓ ✓ nRMSE, MedAE
and the uncertainty
of buildings’ loads
Ribeiro et al. (2018) transfer—based MLP, Long-term Residential 2020 TL-based trend and ✓ MSE
SVR seasonal adjustments
to predict cross-
building load
Ahmad et al. (2020) GSD-GPRM, RBDT, Short-, long-term Office 2020 Building load predic- ✓ ✓ CV
BBRT, BMCDT tion in non-climate
sensitive and climate-
AI‑big data analytics for building automation and management…

sensitiveconditions
Moon et al. (2018) ANN, SVR, PCA-FA Short-term Academic 2020 Energy prediction of ✓ ✓ ✓
higher educational
institutions
Zhang et al. (2020) LSTM, GRU, CIFG Short-term Public 2020 Hybrid DL-based ✓ ✓ CV-RMSE, R2
energy prediction
combined with an
interpretation process
Wen et al. (2020) RNN-GRU​ Short-, mid-term Residential 2020 Achieve well perfor- ✓ ✓ ✓
mance with limited
input variables
Park et al. (2020) XGBoost, RF, DNN Short-term Industrial 2020 A Two-Stage energy ✓ CVRMSE
consumption predic-
tion
4959

13
 Table 3  (continued)
4960

Ref. AI model Forecasthorizon Building nature Year Description Evaluation metrics

RMSE MAE MAPE Others

13
Khamma et al. (2020) GAMs Short-term Office 2020 Embed domain CVRMSE, NMBE
knowledge and prior
understanding of
buildings into the
prediction model
Somu et al. (2020) ISCOA-LSTM Short- , mid-, long- Residential 2020 Accurate and reliabale ✓ ✓ ✓ MSE,Theil U1, U2
term data driven load
forecasting
Liu et al. (2020) A3C, DDPG, RDPG Mid-, long-term N/A 2020 Improve the forecast- ✓ ✓ R2, CV
ing accuracy with
increasing computa-
tion time
Zhang et al. (2020) DBN-DEEM Short-term Residential 2020 Predict stochastic ✓ ✓ ✓ r
energy consumption
using Cyclic feature
(CF) extracted via
spectrum analysis
Lu et al. (2020) CEEMDAN-XGBoost Short-term Intake towers 2020 Have half of prediction RMSPE, Theil U1, U2
error of XGBoost
using real-world
data for a period of
8 years
Wang et al. (2020) stacking model Short-term Academic 2020 Building load forecast- ✓ ✓ ✓ CVRMSE
ing using model
integration
Somu et al. (2021) 𝜅CNN-LSTM Long-term Academic 2021 Capture the load ✓ ✓ ✓ MSE
spatio-temporal
features and aid in
decision making
Y. Himeur et al.
 Table 3  (continued)
Ref. AI model Forecasthorizon Building nature Year Description Evaluation metrics
RMSE MAE MAPE Others

Yuan et al. (2020) WNN-cuckoo search Mid-term Commercial 2020 Optimally tuning the DMAPE, AE
WNN parameters
CS with
2020 a real-world validation
Mawson and Hughes DFNN, RNN Mid-term Industrial 2020 Load forecasting and ✓ ✓
(2020) condition monitor-
ing in manufacturing
buildings
Bui et al. (2021) LSTM Long-, and short-term Residential 2021 Multi-behavior with ✓ ✓ ✓ NRMSE
bottleneck features
LSTM for to predict
energy consumption
Dun and Wu (2020) Grey model Long-term Residential 2020 Load forecasting of ✓
three kinds of build-
ings, i.e. rural, public
AI‑big data analytics for building automation and management…

and urban buildings

Khan et al. (2021) LSTM-KF Short-term Residential 2021 Learning to statistical ✓ ✓ ✓
model for ensemble
predicting of energy
consumption
Li et al. (2021) TL-based ANN Short-term Residential Load prediction of ✓ NTR
information-poor
buildings
Grolinger et al. (2016) NN-SVR Short-term Sport-venues 2016 Load forecasting in a ✓
challenging scenario
with high variations
caused by the hosted
events
4961

13
 Table 3  (continued)
4962

Ref. AI model Forecasthorizon Building nature Year Description Evaluation metrics

RMSE MAE MAPE Others

13
Pinto et al. (2021) RF, GBR Short-term Office 2021 Combine multiple ✓
learners to optimize
the learning process
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4963

energy consumption and temperature in a room of a real-life apartment. In Yang and Wan
(2022), Yang et al., RNN-NARX-based MPC is introduced with instantaneous lineariza-
tion for ACMV optimization. Table 4 summarizes some of the recent ML-based MPC
frameworks described above, their characteristics, and their contributions. Most concen-
trate on controlling HVAC and ACMV systems in buildings since they consume the most
significant proportion of building energy. Thus, this significantly impacts the thermal com-
fort of buildings’ occupants.

3.5.5 Anomaly and fault detection and diagnosis

Failures in building electric networks and devices’ operation cycles may result in exces-
sive energy losses and extra costs. To alleviate these issues, AI-big data analytics are a
prevalent tool that enables detecting faults and disturbances early enough and predicting
maintenance. That is possible by implementing continuous energy consumption monitor-
ing to create ”an early warning system” empowered with AI-big data analytics strategies
and pattern recognition models to notify the end-users and operators Alsalemi et al. (2020).
Accordingly, information related to energy consumption, environmental conditions, and
occupancy patterns is fed into the AI black-boxes to identify and classify deviations. Once
deviations are classified, their causes are determined before taking appropriate measures
for their prevention Himeur et al. (2020), Alsalemi et al. (2020).
In this regard, a plethora of AI-based frameworks have been proposed to develop AI-
big data analytics platforms that allow building energy efficiency Himeur et al. (2021).
In Himeur et al. (2020), a DNN model and the micro-moment concept have been used
to identify energy consumption deviations. A micro-moment rule-based algorithm is
employed to extract load features of daily intent-driven energy usage moments. Next,
DNN is applied to classify and determine abnormal consumption classes automatically
and then compare the performance with various scenarios using conventional ML clas-
sifiers, e.g. LR, LDA, NB, SVM, RF, KNN, DT, ensemble classifier, and MLP. Simi-
larly, in Himeur et al. (2021), unsupervised and supervised anomaly detection schemes
are introduced to promote energy saving in academic and residential buildings. OCSVM
is applied to extract abnormal energy consumption patterns from unlabeled data, while
an improved kNN classifier is proposed to process annotated consumption footprints
that are benchmarked using the micro-moment concept. In Xu and Chen (2020), a
hybrid model using RNNs with quantile regression (QR) is proposed for anomaly detec-
tion in residential houses towards improving the performance of the building and reduc-
ing energy waste (Table 5).
Additionally, fault detection and diagnosis in HVAC systems, being the most exten-
sively operated equipment, has been covered widely in the literature. For instance, the uti-
lization of the different configurations of deep RNNs is investigated in Taheri et al. (2021)
to perform fault detection and diagnosis of common HVAC system faults, such as the mal-
function and leakage of valves and dampers and sensor bias faults. A comparative study is
presented comparing the performance of a DRNN-based diagnosis approach with RF and
GB algorithms. In Yun et al. (2021), a neural network-based supervised auto-encoder (NN-
SAE) - which is an auto-encoder with two outputs that are the classification label and the
reconstructed signal- is proposed for air handling units fault detection and diagnosis before
its validation using ASHRAE experimental data. The two outputs of the NN-SAE have
been then processed to determine the diagnosis decision reliability. This approach has been

13
 4964

13
Table 4  A summary of MPC-based frameworks, their characteristics and contributions in saving energy and optimizing thermal comfort
Work ML model Building type System Best performance Control objective

Gao et al. (2020) FNN + DDPGs Public buildings HVAC 4.31%HVAC energy saving Save energy and improve thermal comfort.
Yang et al. (2020) ANN-NARX Office and lecture theater ACMV 58.5% cooling thermal saving Energy saving and thermal comfort optimi-
zation.
Yang et al. (2021) RNN-NARX Office and a lecture theater ACMV 52% reduction of cooling energy Save energy and optimize thermal comfort.
in experimental testbeds
Chen et al. (2020) MLP-based transfer learning Residential buildings HVAC MSE=0.16 Optimize energy efficiency and thermal
comfort.
Bünning et al. (2020) RF Residential buildings HVAC 24.9% of cooling energy saving Optimize energy consumption withoutcom-
promising thermal comfort
Yang and Wan (2022) RNN-NARX Office in a hospital ACMV 26–31.6% cooling energy savings Save energy and optimize thermal comfort
Li and Tong (2021) Encoder-decoder RNN Residential/public buildings HVAC 4–7% energy saving Energy saving and smart control of thermal
environment
Mtibaa et al. (2021) CAM- LSTM Multi-zone buildings HVAC MAPE = 0.0872% Save energy, predict peak power and improve
thermal comfort
Y. Himeur et al.
 Table 5  A summary of AI-Big data analytics models used for anomaly and fault detection and diagnosis
Ref. AI model Fault System Targeted system Year Description Evaluation metrics
anomaly
Sensor Component Actuator ACC​ MAE MSE Others

Himeur et al. DNN ✓ Building appli- 2020 Classify and ✓ Precision, F1-score,
(2020) ances determine AUROC curves,
abnormal confusion matrix
building energy
consumption.
Himeur et al. OCSVM, kNN ✓ Building appli- 2021 Classify and ✓ F1-score
(2021) ances determine
abnormal
building energy
consumption.
Xu and Chen RNN, QR ✓ House 2020 Detect abnormal PICP, PINAW
(2020) energy con-
sumption.
Taheri et al. RNN ✓ ✓ ✓ HVAC systems 2021 Investigate ✓ Precision, recall,
(2021) various DRNN F1-score, cross
AI‑big data analytics for building automation and management…

configurations entropy, confu-

to perform sion matrix
HVAC system
fault diagnosis.
Yun et al. (2021) NN-SAE ✓ ✓ AHUs 2021 Detect and diag- Precision, recall,
nose common F1-score
AHUs faults to
enable reliable
maintenance.
4965

13
 Table 5  (continued)
4966

Ref. AI model Fault System Targeted system Year Description Evaluation metrics
anomaly
Sensor Component Actuator ACC​ MAE MSE Others

13
Li et al. (2021) GAN ✓ HVAC systems 2021 Detect and diag- ✓ FPR
nose common
HVAC system
faults and
addressing the
issues of labeled
data availability
and data imbal-
ance.
Elnour et al. AANN ✓ HVAC systems 2020 Perform sensor ✓ TPR, FPR, recovery
(2020) data validation rate, deviation
and fault diag- rate, noise reduc-
nosis of HVAC tion rate
system sensor
faults using
semi-supervised
learning.
Elnour and 2D CNN ✓ HVAC systems 2021 Diagnose HVAC ✓ Precision, recall,
Meskin system single F1-score, speci-
actuator faults ficity
using supervised
learning.
Dey et al. (2020) MC-SVM ✓ ✓ HVAC system 2020 Provide auto- Precision, recall
mated detection
and diagnosis
of equipment
failures in
HVAC systems’
terminal units.
Y. Himeur et al.
 Table 5  (continued)
Ref. AI model Fault System Targeted system Year Description Evaluation metrics
anomaly
Sensor Component Actuator ACC​ MAE MSE Others

Bode et al. (2020) LR, kNN, ✓ Heat pumps 2020 Investigate fault ✓ MCC
CART,RF, NB, detection
SVM, NN approaches for
operational heat
pumps using
machine learn-
ing algorithms.
Shahnazari et al. RNN ✓ ✓ HVAC systems 2019 Develop models ✓ ✓ ✓
(2019) and a fault diag-
nosis methodol-
ogy for HVAC
systems.
Dey et al. (2020) Clustering ✓ ✓ FCUs 2020 Remote detection Silhouette indexing,
of fan coil units Davies-Bouldin,
common faults. maximal gap
AI‑big data analytics for building automation and management…

Han et al. (2019) LS-SVM ✓ ✓ Chiller system 2019 Diagnose com- ✓

mon faults in
centrifugal
chillers.
Liu et al. (2021) CNN ✓ ✓ Chiller system 2021 Assess transfer ✓ Recall, F1-score,
learning for precision confu-
fault diagnosis sion matrix
methods in
chillers.
Zhu et al. (2021) DANN ✓ ✓ Chiller system 2021 Apply transfer ✓ Accuracy improve-
learning for ment degree
chiller fault
diagnosis.
4967

13
 Table 5  (continued)
4968

Ref. AI model Fault System Targeted system Year Description Evaluation metrics
anomaly
Sensor Component Actuator ACC​ MAE MSE Others

13
Han et al. (2020) kNN, SVM, RF ✓ ✓ Chiller system 2020 Assess using ✓ F1-score, confusion
ensemble learn- matrix
ing for chiller
fault diagnosis.
Choi and Yoon NN-AE ✓ ✓ BAMS 2021 Investigate ✓ Precision, F1-score
(2021) variants of
the AE-based
fault diagnosis
approach for
building auto-
mation systems.
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4969

compared with conventional ANN and SVM algorithms, and it has been found reliable as it
considers undefined situations.
Additionally, a 2D CNN-based HVAC system actuator fault diagnosis is introduced in
Elnour and Meskin, in which the system’s measurements and control signals were con-
figured into multi-channel images and then processed using the 2D CNN-based diagno-
sis framework. CNNs are characterized by their high-performance accuracy and powerful
capability in learning and realizing complex functions and interdependency from any given
data. While in Liu et al. (2021), a CNN-based chiller fault diagnosis method is developed
for building energy systems. Additionally, a TRL-based scheme is assessed to investigate
the potential of using a pre-trained CNN-based fault diagnosis approach for chillers with
different specifications, which is useful when available data is limited in size and/or types
of operating conditions/faults captured.
In Dey et al. (2020), a big-data framework is presented to enable automated HVAC sys-
tem fault diagnosis in large scale buildings in which a feature extraction approach is pro-
posed to reduce the data dimensionality, and a multi-class SVM (MCSVM) algorithm is
utilized to develop the diagnosis model. It aims to provide energy savings through preemp-
tive maintenance, behavior analysis, and predictive building identification. In Bode et al.
(2020), various ML algorithms are investigated to perform fault detection on a heat pump
system, which are LR, kNN, classification and regression tree (CART), RF, NB, SVM, and
NNs. This study demonstrates the effect of the data quality and amount and the limitations
on the system’s features availability (i.e, types of available sensors) on the performance of
the developed ML models. While in Han et al. (2020), a supervised hybrid fault diagnosis
model using SVM, kNN, and RF is developed for chiller fault diagnosis such that the three
models are developed independently to perform fault diagnosis; then the final decision is
made based on the plurality voting method. It was found that ensemble learning contributes
to diagnostic performance improvements. In Li et al. (2021), a fault diagnosis approach
is proposed for the common component faults in the HVAC system of an office building
using a modified GAN. The proposed approach enables leveraging the labeled and unla-
beled data simultaneously such that it aims to process the unlabeled data and utilize the
limited information from the labeled ones to conclude the diagnosis decision.
Those fault diagnosis approaches mainly require sufficient labeled data for training,
which can be unavailable or complex, and costly to obtain. Therefore, several studies have
developed unsupervised and semi-supervised diagnosis strategies as in Elnour et al. (2020),
where an auto-associative neural network (AANN) is utilized for sensor data validation
and fault diagnosis in HVAC systems using semi-supervised learning. It demonstrates a
compelling performance in sensor error correction, data replacement of unavailable sen-
sors, measurement noise reduction, and sensor inaccuracy correction. Also, it is effective
for both single and multiple sensor faults diagnoses. In Zhu et al. (2021), transfer learning
is applied to develop a chiller fault diagnosis approach that only requires the system’s nor-
mal operation data using domain adversarial neural network (DANN). While in Shahnaz-
ari et al. (2019), a distributed diagnosis approach using RNNs utilizing the normal system
operation data is developed for multiple sensors and actuator faults diagnosis. It is based on
developing intercommunicating fault detection and isolation (LFDI) agents for the various
HVAC subsystems, i.e., cooling coil, VAV box, etc., and each LFDI agent is composed of
two RNN-based models. It demonstrates promising capability in fault diagnosis. However,
it is excessively computationally demanding, given the two RNN-based models included in
each agent.
Additionally, a multi-level automatic fault detection framework is proposed in Dey
et al. (2020) for fan coil units (FCUs). Feature extraction followed by data clustering are

13
4970 Y. Himeur et al.

applied to identify faulty and healthy data, and then a clustering-based fault diagnosis
model is developed. The least-squares support vector machine (LS-SVM) regression model
is used in Han et al. (2019) to develop a chiller fault diagnosis strategy that is validated
using ASHRAE data. The proposed approach is compared with two other methods using
the SVM algorithm of probabilistic neural networks (PNNs). In Choi and Yoon (2021),
a semi-supervised fault diagnosis approach is proposed for building automation systems
using NN-based auto-encoders (AEs). An AE is a structure that transfers the input to the
latent space then uses the compressed representation to produce a reconstructed version at
the output. Variants of the proposed method are investigated: the residual-based approach
using the error between the original and the reconstructed signal as the indication of the
system status, and the latest space-based approach in which the features of the compressed
representation are used for fault diagnosis.

3.5.6 Indoor environmental quality (IEQ) monitoring

IEQ monitoring continues to grow in importance, several works have demonstrated an

apparent relationship between the increasing concentration of CO2 and decreasing cogni-
tive performance (Nejat et al. 2020; Pulimeno et al. 2020). Typically, monitoring ambient
IEQ and temperature can reveal valuable information for creating a healthier, more com-
fortable environment for end-users. It is also a prime opportunity for energy and cost sav-
ings (Saini et al. 2020a). Thus it becomes possible, using detailed AI analytics in BAMSs,
to identify various environmental problems, including air pollution, where different pollut-
ants may affect the IEQ , such as the cleaning products, cigarettes smoke, perfumes, con-
struction activities, water-damaged building materials, and other types of outdoor pollut-
ants (Saini et al. 2020b). Indeed, albeit these gazes are commonly safe for end-users, their
effect on human health can be dangerous if they exceed certain thresholds of exposure. To
that end, an intelligent IEQ monitoring system for classifying and recognizing diverse pol-
lutants and measuring their levels is of utmost importance (Wei et al. 2019; Muiruri et al.
2021).
Before the COVID-19 outbreak, IEQ monitoring was not a priority in public buildings,
e.g. sport venues, banks, healthcare centers, academic institutions, commercial centers, res-
taurants, and so on. However, the fast proliferation of the corona virus and its resulting
harmful effects have put IEQ in the spotlight as an important component of BAMSs. In
Mumtaz et al. (2021), the authors (i) develop an IoT node including various sensors for
collecting data, (ii) introduce a NN model for classifying 8 pollutants, and (iii) design an
LSTM-based DL model for predicting the concentration of every pollutant and the overall
IEQ. In Mad Saad et al. (2017), a pollutant recognition scheme is proposed for IEQ moni-
toring using different supervised ML algorithms, including MLP, KNN and linear discrim-
ination analysis (LDA). The evaluation has been conducted in a residential building located
in a rural area in China, where 5 different indoor air pollutants were considered (combus-
tion activity, presence of chemicals, presence of food and beverages, ambient air, and pres-
ence of fragrances). While in Loy-Benitez et al. (2020b), Loy et al. introduce an ML-based
scheme for detecting, diagnosing, identifying, and reconstructing abnormal observations
of multivariate IEQ data in a subway station. Accordingly, a memory-gated RNN-based
autoencoders (MG-RNN-AE) that can process dynamic and sequential IEQ data has been
utilized.
In Cruz et al. (2020), an IEQ prediction model is developed using SVM radial basis func-
tion (SVM-RBF) and stochastic Gradient Boosting machines (SGBM). The performance of

13
AI‑big data analytics for building automation and management… 4971

of these models has been evaluated using root mean squared error (RMSE) and R2, and a
comparison with other ML algorithms has been presented. In Taştan and Gökozan (2019),
a real-time IEQ monitoring system is designed using and IoT-based e-nose and diverse ML
classifiers, including SVR, generalized regression neural network (GRNN), and extreme
learning machine (ELM) with Gaussian kernels. The linear correlation (LC) has been used
for evaluating this framework. In Sharma et al. (2021), a cost-effective framework for IEQ
prediction is introduced, where MLP and eXtream Gradient Boosting Regression (XGBR)
are used for providing real-time measurements of the concentration of pollutants, i.e. CO2
and particulate matter 2.5 (PM2.5) in a set of classrooms. Moving on, an LSTM without
using the forget gate (LSTM-wF) is deployed to predict the air quality at a lower complex-
ity and increase the prediction performance.
In Alawadi et al. (2020), an indoor temperature forecasting scheme is proposed, where
up to 36 ML models (pertaining to 20 different families) have been deployed. Real-world
data gathered for three hours from both smart households and weather station have been
used to validate this study. Similarly, in Aliberti et al. (2019), Aliberti et al. propose a
smart solution for indoor air-temperature prediction, where a non-linear autoregressive
neural network (NN-ARNN) has been utilized to perform short- and medium-term fore-
casting. This model has been then validated on both a synthetic dataset and real-world data
recorded using IoT devices installed in residential buildings.
As CO2 concentration is appropriate for measuring the IEQ quality due to its over the
sensor networks, a set of frameworks have adopted it. For instance, in Taheri and Razban
(2021), an ML-based IEQ monitoring approach is proposed by predicting CO2 concentra-
tion in the academic building (campus classrooms) using demand-controlled ventilation.
Various ML algorithms have been employed and compared to learn the CO2 concentration,
among them SVM, AdaBoost, RF, Gradient Boosting (GB), LR, and MLP. In a similar
way, Kallio et al. (2021) propose a smart approach to forecast office indoor CO2 concentra-
tion by adopting four ML algorithms, i.e. ridge regression (RR), DT, RF, and MLP. Moreo-
ver, a baseline to evaluate the indoor CO2 prediction has been introduced by producing
a benchmark dataset covering an entire year. Moving on, In Tagliabue et al. (2021), an
ML-based IEQ monitoring approach that relies on measuring the CO2 concentration in an
academic building is proposed. Specifically, an LSTM-based RNN models and IoT sen-
sors have been then used for monitoring the indoor conditions depending on the occupancy
patterns.
Other IEQ monitoring frameworks have focused on measuring and predicting other
factors, which are recommended in various countries, such as total volatile organic com-
pounds (TVOC), formaldehyde (HCHO), and carbon monoxide (CO). For instance, in
Chen et al. (2018), Chen et al. use four ML models, including SVM, Gaussian processes
(GP), M5P and backpropagation neural network (BPNN) for predicting CO2, HCHO, and
TVOC in an academic building (in Singapore). In a similar way, in Lagesse et al. (2020),
various ML models are utilized for predicting PM2.5 in office buildings, i.e. ANN, LSTM,
multiple linear regression (MLR), partial least squares regression (PLS), distributed lag
model (DLM), and least absolute shrinkage selector operator (LASSO).
Other AI-big data analytics have also been used to perform additional tasks. For exam-
ple, Loy et al. (2020a) introduce a variational autoencoder (VAE) coupled with convo-
lutional layers (VAE-CNN) model to impute missing IEQ data. Accordingly, two sce-
narios have been adopted to evaluate the VAE-CNN algorithms: (i) a point-to-point data
removal, and (ii) data intervals removing at different sampling rates. While in Kalajdjieski
et al. (2020), the capability of generative adversarial networks (GANs) is exploited in

13
4972 Y. Himeur et al.

combination with a data augmentation technique for overcoming the class-imbalance issue
while monitoring IEQ using large-scale datasets. Table 6 outlines pertinent AI-Big data
analytics frameworks introduced for monitoring IEQ and performs a comparison between
them, with reference to the AI model, forecast horizon, building environment, year of
appearance, method description and evaluation metrics.

3.5.7 Security and safety

Among the major safety concerns in general and in buildings, in particular, are the fire out-
break. The prevention of and the immediate reaction to fires minimizes their consequences
in terms of the people’s well-being and the financial losses. This includes the minimization
of fire incidents potentials, the fast detection and extinguishment of the fire source, the
effective execution of emergency evacuation, and the prompt notification of the emergency
situation to the concerned authorities. Buildings are usually equipped with conventional
fire alarm and extinguishment systems consisting of several sensing devices, including
smoke, heat, and flame detectors, automated alarms, and water sprinklers (Zverovich et al.
2016). With the advent of big data algorithms and analytics, the fire safety in buildings can
be boosted by employing buildings data to develop frameworks for fire prevention, detec-
tion, and suspension. An analysis of the advantage of utilizing ML algorithms for reliable
and prompt fire detection was provided in Surya (2017) represented in their distinguished
capability of black-box modeling, feature extraction, pattern recognition with high accu-
racy and reliability. Unlike conventional fire detection systems, ML-based models can be
used to detect fire, analyze it effects, assess its risks, predict its behavior utilizing the data
collected from the sensing devices.
For example, in Zhang et al. (2021), a model combining a deep belief network (DBN)
and a recurrent LSTM neural network (R-LSTM-NN) was proposed for fire hazards pre-
diction in smart cities. The proposed model was used in predicting the air quality that is
then used to detect fire outbreaks based on the sensors readings of the IoT system. It shows
promising potential in when data records of the IoT system are available for normal opera-
tion and scenarios of fire occurrence. In Fu (2020), a comparative study was presented
using ML algorithms, namely, DT, KNN, and NNs, to develop classification models to pre-
dict failure patterns and to assess the progressive collapse potential for steel framed build-
ings in fire. The study aimed to develop a reliable fire assessment tool for practitioners and
the developed framework demonstrated a satisfactory performance overly. In Sultan Mah-
mud et al. (2017), another comparative study was conducted for developing an intelligent
fire detection system with early notification system in which data mining algorithms such
as DTs, Bayesian networks (BayesNet), NNs, and SVM were used to develop data-driven
classifiers using supervised learning. Moreover, the proposed smart fire detection system
employed an edge detection model to analyze the data collected from the cameras to con-
firm the fire detection decision. In Huda et al. (2012), an AI-based framework was pro-
posed to assess the thermal condition of electrical installations in buildings to prevent the
potential of injuries and fire hazards using infrared images. Raw data was processed using
PCA for features selection that are then used to develop an NN-based classifier to deter-
mine the condition of electrical equipment.
In Ouache et al. (2021), a fire safety assessment framework was proposed to help pre-
dict the potential fire impacts and recommending optimal fire intervention strategies in
multi-unit residential buildings using NNs. Supervised learning was used to train a NN
based on 5 predictors, among which are the mean of the initial fire detection (i.e., smoke

13
 Table 6  A summary of AI-Big data analtiycs frameworks introduced for IEQ monitoring
Ref. AI model Task Building nature Year Description Evaluation metrics Others
RMSE MAE MAPE ACC​ F1

Mad Saad et al. KNN, LDA, MLP Pollutant ecognition Residential 2017 Recognize 5 different ✓
(2017) pollutants using
supervisedlearning
Cruz et al. (2020) SVM-RBF, IEQ prediction Academic (labs) 2020 Forecast IEQ and the ✓ R2
ideal number of
occupants in a lab
Alawadi et al. (2020) 36 ML models Temperature predic- Residential 2020 Compare a set of 36 ✓ R2
tion ML algorithms for
temperature predic-
tion
Moon et al. (2018) SVR, GRNN, ELM IEQ monitoring Residential 2018 Real-Time monitor- LC
ing of IEQ with an
IOT-based e-nose
Taheri and Razban SVM, RF, LR, GB CO 2 concentration Academic 2021 Predict CO2 con- ✓ ✓ ✓ ✓ R2
(2021) MLP, AdaBoost prediction centration with
AI‑big data analytics for building automation and management…

the ability to learn

nonlinearities con-
nected with the CO2
data
Loy-Benitez et al. MG-RNN-AE IEQ monitoring Subway station 2020 Detect, diagnose, ✓ MSE, R2
(2020b) identify, and recon-
struct abnormal
observations of
multivariate IEQ
data
4973

13
 Table 6  (continued)
4974

Ref. AI model Task Building nature Year Description Evaluation metrics Others
RMSE MAE MAPE ACC​ F1

13
Kallio et al. (2021) RR, DT, RF, MLP CO 2 concentration Residential 2021 Forecast office indoor ✓ ✓
prediction CO2concentration
and introduce a
benchmark dataset
that covers a full
year.
Tagliabue et al. RNN IEQ prediction Academic Monitor the ventila- MSE, R2
(2021) tion rate using an
IoT protocol by
analyzing data with
ANN communica-
tion
Aliberti et al. (2019) ARNN Temperature predic- Residential 2019 Predict indoor air- MAD, RMSD
tion temperature using
synthetic data to
train an ARNN
Chen et al. (2018) SVM, GP, M5P, Prediction of CO 2 , Academic 2018 Analyze CO2, ✓ R2
BPNN TVOC and HCHO TVOC and HCHO
footprints for better
predicting the IEQ
Sharma et al. (2021) MLP, XGBRLSTM- IEQ monitoring - 2021 Estimate and predict ✓ ✓ ✓
wF the concentration of
pollutants i.e. CO 2
and MP 2.5
Lagesse et al. (2020) ANN, LSTM, MLR Prediction of PM2.5 Commercial 2020 Predict PM2.5 for ✓
PLS, DLM, LASSO IEQ monitoring in
commercial office
buildings
Y. Himeur et al.
 Table 6  (continued)
Ref. AI model Task Building nature Year Description Evaluation metrics Others
RMSE MAE MAPE ACC​ F1

Mumtaz et al. (2021) NN, LSTM Classification + Public 2021 Detect anomalies of ✓ ✓ ✓ ✓ MSE
Prediction IEQ and predict the
concentration of
each air pollutant
AI‑big data analytics for building automation and management…
4975

13
4976 Y. Himeur et al.

detector, heat detector, visual, etc.), the action taken to fight fire (i.e., occupant response,
fire department, BMS, etc.), and the performance of the BMS in fire detection and exten-
sion. The fire impact assessment covered several aspects, which are the occupant response
to the incident, fire extension, fire damage, and financial losses. The proposed framework
demonstrated a remarkable ability to predict fire impacts accurately, and it represented a
promising solution to define and regulate fire safety strategies.
The security of the automation and management system in buildings has become more
imperative due to the rapid advancement in the technologies used and the IoT systems.
The industry predicts that the IoT market will grow from an installed base of 30.7 Bil-
lion devices in 2020 to 75.4 Billion in 2025 (IoT Security Foundation), which will expose
them to increased risk of advanced attack vectors. According to Kaspersky Lab, nearly four
in ten buildings were targeted by attacks in the first half of 2019 , and it is expected that
the impact of cyberattacks on the building and construction industry will be significant
in the coming years (Kaspersky). In Elnour et al. (2021), an attack detection framework
for false data injection was proposed for a multi-zone HVAC system in office buildings
utilizing an isolation forest (IF) algorithm. The operational data of the system’s sensor and
control command signals were used to develop the detection model using semi-supervised
learning. Isolation forests are characterized by the low computational requirement and
capability to handle to complex and multi-variate data. They work based on pointing out
anomalies using the concept of isolation, which improves the attack detection capability.
Feature selection was applied to the raw data, and the study presented a comparative analy-
sis of two models for feature reduction, which are PCA based model and a 1D-CNN-based
model.
In de Assis et al. (2020), a security system for industrial IoT was proposed in which
a CNN-based classifier was developed to identify distributed denial of service (DDoS)
attacks in software-defined networks (SDNs). This system is based on supervised learn-
ing from the labeled network data of the IoT system. CNNs are advantageous for their
high accuracy and classification performance, and powerful capability in realizing com-
plex interdependency from multi-variate and sophisticated data. While in Aboelwafa et al.
(2020), a residual-based attack detection framework was presented in which an NN-based
auto-encoder was trained to profile normal system behavior. Then, non-conforming obser-
vations are identified as anomalies based on the generated residuals between the input and
the output of the AE. Auto-encoders are used to learn the latent feature representation
of the system using healthy operational data. They are also used for data dimensionality
reduction, noise filtering, information retrieval, etc. In Yahyaoui et al. (2020), a preliminary
demonstration of a ML-based intrusion detection system for data protection in healthcare
centers was presented. An SVM model was developed using the labeled data of the IoT net-
work and it demonstrated a promising performance in detecting malicious actions launched
against the IoT system. The accuracy of the proposed framework was assessed based on
the energy consumption in the communication network because attacks result in increased
energy usage due to the increased network traffic. Table 7 highlights and compares existing
AI-Big data analytics frameworks introduced to ensure security and safety in BAMSs.

3.5.8 Occupancy detection

Occupancy data are collected by various sensors and devices in buildings to help improve
the efficiency of the BAMSs in terms of energy utilization and occupants’ comfort and
well-being. These include cameras, infrared sensors, and carbon dioxide detectors

13
 Table 7  A summary of AI-Big data analytics frameworks proposed to ensure security and safety applications
Ref. AI model Task Building nature Year Description Evaluation metrics
ACC​ MAE MSE Others

Zhang et al. (2021) DBN, R-LSTM-NN Fire hazard prediction Smart cities 2021 Detect fire outbreak ✓ TPR, FPR Error rate
based on the sensors
readings of the IoT
system using super-
vised learning
Fu (2020) DT, kNN, NN Fire safety assessment Steel framed build- 2020 Predict failure pat- ✓
ings terns and pro-
gressive collapse
potential due to fire
Sultan Mahmud et al. DT, BayesNet NNs, Fire detection - 2017 Enhance the fire ✓ ✓ ACC, Precision RAE,
(2017) SVM system detec- Recall, F1
tion capability by
analyzing system
data to conclude the
situation
AI‑big data analytics for building automation and management…

Huda et al. (2012) PCA, NN Inspection of electri- Office 2012 Assess the thermal ✓
cal installations condition of electri-
cal installations
based on infrared
images using super-
vised learning
Ouache et al. (2021) NN Fire safety assessment Residential 2021 Investigate and assess R
fire protection and
intervention strate-
gies
Elnour et al. (2021) IF, PCA, CNN Attack detection Office 2021 Detect attacks on the Precision, Recall
building manage-
ment system
4977

13
 Table 7  (continued)
4978

Ref. AI model Task Building nature Year Description Evaluation metrics

ACC​ MAE MSE Others

13
de Assis et al. (2020) CNN DDoS attack detection Industrial 2020 Detect DDoS attacks ✓ Precision, Recall F1
on the SDNs of the score
IIoT system
Aboelwafa et al. NN FDI attack detection Industrial 2020 Detect FDI attacks on ✓ ✓ FPR, TPR
(2020) IIoT systems
Yahyaoui et al. (2020) SVM Intrusion detection Hospital 2020 Detect intrusion in Energy
IoT systems
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4979

(Sardianos et al. 2020; Sayed et al. 2022). The data can be directly used as inputs to con-
trol and regulate some of the buildings’ equipment, such as lights, air conditioning, doors,
etc. Additionally, scholars and researchers utilize big data analytics to develop approaches
for analyzing and processing building occupancy data to facilitate an efficient and reliable
overall building management (Sardianos et al. 2020).
In Huang and Hao (2020), a DL-based visual recognition was used to implement an
occupancy detection framework utilizing CNNs. The proposed approach was used to deter-
mine the number of people present and their location to help operate demand-based HVAC
systems more reliably and efficiently. It was found that the proposed approach outperforms
the conventional occupancy detection systems in terms of accuracy, precision, robustness,
ability to provide occupancy count, and ability of static and dynamic occupancy detection.
It also requires hardware and computational considerations as CNNs have a high computa-
tional overhead. In Acquaah et al. (2020), a study was presented to estimate the occupancy
count based on thermal images using CNNs for feature extraction and SVM for multi-class
classification. Two well-known CNN architectures were investigated, which are the 50 lay-
ers ResNet (ResNet-50) and AlexNet using transfer learning due to their superior perfor-
mance in image processing (He et al. 2016; Krizhevsky et al. 2012). In Tien et al. (2021), a
computer vision-based occupancy and equipment usage detection framework was proposed
to facilitate a demand driven control of the HVAC system in an office room. A multi-class
region-based CNN model was developed and deployed to analyze camera images. It can
predict the occupancy count, activity type (i.e. sitting, walking, etc.), and equipment usage
in real-time.
A thermal-based occupancy detection approach was presented in Zhao et al. (2018)
in which data-driven models were developed using SVR and RNN to predict occupancy
information using the building’s properties, including indoor temperature, towards man-
aging energy use and security monitoring in intelligent buildings. That is, the interac-
tion of the thermal components present in the conditioned space and the necessary part
of determining thermal consistency is indicated by the indoor temperature. Supervised
learning was used to train the ML models using simulation data generated from Energy
Plus such that the target outputs were the occupancy count. The work in Elkhoukhi
et al. (2020) combined the IoT technology and Big data analytics to implement real-
time occupancy detection such that data of the indoor lighting, temperature, humidity,
and CO2 levels were used to predict the status of the building occupancy. Two models
were developed and tested, one using LDA and the other using vertical hoeffding tree
(VHT) for offline and online occupancy detection, respectively. Additionally, in Fatema
and Malik (2021), feature extraction and correlation analysis were performed on indoor
sensors data (i.e., temperature, CO2 level, light intensity, humidity) and then particle
swarm optimization was used to train an NN-based occupancy detection model. It dem-
onstrated improved classification performance compared to conventional NNs optimized
using the back-propagation algorithm.
In Wu and Wang, a ML-based model was proposed to improve the operation of the
BAMS due to the shortcomings of infrared sensors for stationary occupancy. The model
predicted the occupancy status based on multiple statistical features of the signals acquired
by the infrared sensors. ML algorithms were investigated, among which SVM demon-
strated the best performance due to its ability to capture complex and nonlinear functions,
and its efficacy in handling high dimensional data. In Huchuk et al. (2019), a comparative
analysis was presented for occupancy forecasting using ML algorithms based on thermo-
stat data. The prediction model was intended to optimize the operation of the air condition-
ing system, such that both the present and the future occupancy information is taken into

13
4980 Y. Himeur et al.

consideration. It was found that RF algorithm outperformed the LR, the Markov model, the
hidden Markov model (HMM) and the RNN, which is based on the bagging technique in
which multiple models on different subsets of the training dataset are developed, then their
predictions are combined to conclude the final output of the RF model. However, occu-
pancy forecasting is only dependable when the building does not exhibit rapid and random
fluctuations in the user profile, which is generally the case for residential buildings.
In Razavi et al. (2019), a comparative study was presented for the utilization of super-
vised ML algorithms such as SVM, RF, KNN, and NNs to estimate and predict occupancy
information in residential buildings based on power meters data. It was found that the
reliability of occupancy prediction is lower for the larger forecast horizons. In Feng et al.
(2020), a DL-based approach is proposed combining a CNN and a bidirectional LSTM
network for occupancy detection in houses based on electrical data of advanced metering
infrastructures (AMIs). The data essentially contain readings of electric current, voltage,
and power that are processed by the CNN for spatial feature extraction. Using supervised
learning, the extracted features are then fed to the BiLSTM network to solve a binary classi-
fication problem to identify the occupancy condition in real-time. The proposed framework
demonstrated improved performance when compared to other ML and DL based models
due to its ability to interpret the spatial and contextual features of the data. However, since
detailed occupancy information are mostly not recorded and hence not available, super-
vised learning-based approaches that are based on such a detailing in the data (i.e., people
count) can be impractical and difficult to implement using actual building data. While in
Pešić et al. (2019), a LSTM network-based framework was proposed to perform occupancy
detection and forecasting as well as data analytics based on Bluetooth positioning and WiFi
utilization data of the IoT infrastructure in a multi-story residential building. The network
data were pre-processed to extract the information of the occupancy of the apartments,
then used to develop the LSTM network to predict and forecast occupancy condition and
patterns in the different spaces of the building. The proposed work demonstrated the effec-
tive fusion of Bluetooth and WiFi data as well as the successful deployment of NN-based
data analytics using wireless networks data for occupancy detection application. Table 8
summarizes the relevant AI-Big data analytics frameworks developed to detect occupancy
profiles.

3.5.9 Water usage management

Almost all kinds of buildings are users of water, although the cost of water and sewer ser-
vices varies from area to area and can become a significant expense. Worse, in areas where
there is a shortage of water, it is not only a big expense, but an imperative to conserve.
Therefore, it becomes of significant importance to bring the monitoring of water levels
and switching points of all wet applications in buildings to the BAMS. Furthermore, water
monitoring systems can benefit from the advancement of AI and ML technologies for
improving their performance (Sun and Scanlon 2019). Typically, by harnessing the power
of AI-big data analytics, it is possible to maximize information and data available and
hence make better decisions while enhancing service delivery and reducing costs (Rahim
et al. 2020).
In this context, using IoT water meters with wireless connectivity (Bluetooth,
LoRaWAN, etc), it becomes relatively easy to install water meters within the building.
These can be as simple as pulse style meters that can easily be integrated into a BAMS.
Moving on, adopting AI-big data analytics to analyze data from water meters has become

13
 Table 8  A summary of AI-Big data analytics models used for occupancy detection
Ref. AI model Detection basis Building nature Year Description Evaluation metrics
ACC​ MAE MSE Others

Huang and Hao (2020) CNN Surveillance cameras Office 2020 Detect the number and ✓ Relative error
location of occupants
Acquaah et al. (2020) CNN, SVM Thermal cameras – 2020 Estimate the number of ✓
people present based
on thermal images
Tien et al. (2021) CNN Vision cameras Office 2021 Predict equipment use ✓ Precision, Recall, F1
and occupancy count
& activity in real
time
Zhao et al. (2018) SVR, RNN Temperature sensor Office 2018 Detect the number Error rate
of occupants based
on indoor thermal
properties
Elkhoukhi et al. (2020) LDA, VHT Indoor sensors Office 2020 Predict the status of ✓
occupants’ presence
AI‑big data analytics for building automation and management…

Fatema and Malik NN Indoor sensors Office 2021 Predict occupancy ✓ TPR, FPR, Precision,
(2021) condition in an office Recall, F1, MCC
room
Wu and Wang SVM, kNN, DT RF, Infrared sensor – 2021 Provide accurate pre- ✓
NN dictions of the occu-
pancy status based on
motion detectors
Huchuk et al. (2019) LR, HMM, MM, RF, Thermostat data Residential 2019 Forecast the occupancy ✓
RNN information
Razavi et al. (2019) kNN, SVM, NN, RF Energy meter Residential 2019 Estimate and predict ✓ Precision, AUROC
occupancy informa-
tion
4981

13
 Table 8  (continued)
4982

Ref. AI model Detection basis Building nature Year Description Evaluation metrics
ACC​ MAE MSE Others

13
Feng et al. (2020) CNN, BiLSTM Smart meters Residential 2020 Predict real-time occu- ✓ Precision, Recall, F1,
pancy status based TNR F1, Training time
on data of electrical
signals
Pešić et al. (2019) LSTM Bluetooth and WiFi Residential 2019 Predict, forecast, and ✓ RMSE and Edit Distance
devices analyze occupancy on Real Signals
information using
wireless networks
data
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4983

Fig. 5  Flowchart of water usage monitoring system used to detect water leaks and optimize water consump-
tion (Jenny et al. 2020)

crucial for optimizing the management of water resources and sustaining growth and devel-
opment. Accordingly, various AI-big data analytics frameworks have recently been pro-
posed with the aim of (i) processing complex nonlinear water data, (ii) forecasting water
demand, (iii) predicting water meter failures, or (iv) monitoring the quality and tempera-
ture of the water. Fig. 5 illustrates the flowchart of a water usage monitoring system used to
detect water leaks and optimize water consumption (Jenny et al. 2020).
In Altunkaynak and Nigussie (2017), water demand prediction is conducted by first using
multiplicative season algorithm (MSA) to extract pertinent information from water meter
records and also capturing periodicity and converting nonstationary signals into stationary sig-
nals. Following, their output is fed into an MLP for accurately predicting water demand. The
RMSE and Nash-Sutcliffe coefficient of efficiency have been adopted to evaluate the predic-
tion performance of the learning model and its ability to extend prediction lead time. Shine
et al. (2018), diverse ML models are used for predicting water consumption in an agricultural
building based on analyzing data collected from a remote monitoring system. Thus, RF, ANN,
SVM, and CART decision tree (CDT) algorithms were trained to predict water consumption,
where a backward sequential variable selection was adopted for excluding variables adding
low predictive power along with a hyper-parameter tuning with nested cross-validation for cal-
culating the prediction accuracy for each model. In a similar way, in Smolak et al. (2020),
three ML algorithms are implemented and compared for predicting water usage, i.e. RF, SVM,
ARIMA. The water consumption data augmented with end-users occupancy patterns were
used to improve the prediction accuracy. A novel approach to process and correlate between
occupancy and water usage time-series was introduced. This framework was validated on 51
days of water consumption readings and over 7 million occupancy patterns from urban areas.
On the other hand, by using AI-big data analytics, it is also possible to monitor water quality
and hence improve water resources management plans. In Chen et al. (2020), 10 ML models
are deployed to water quality prediction (WQP). Specifically, DT, NB, LR, LDA, completely-
random tree (CRT), KNN, SVM, RF, and deep cascade forest (DCRF) have been trained using
water data from a hydro-electric power (HEP) plant, including pH, DO, CODMn, and NH3-N
to forecast water quality. The precision, recall, F1 score, and weighed F1-score (wF1) have
been selected to evaluate the prediction performance of the ML algorithms. In Roccetti et al.

13
4984 Y. Himeur et al.

(2019), Roccetti et al. develop an ML-based classifier, which is personalized for predicting the
failure of a water meter. Typically, an RNN model is deployed for (i) processing 15 million
of readings collected from 1 million of mechanical water meters, and (ii) extracting relevant
patterns representing the complex phenomenon of defective water meters. This has helped in
achieving more than 80% accuracy in detecting failures.
In Wang et al. (2018), water demand of urban areas is predicted using gravitational search
algorithm (GSA) and backtracking search algorithm (BSA) with ANN with regard to various
weather parameters. While in Antunes et al. (2018), four ML models are selected to predict
water demand, including ANN, RF, SVM and KNN, through the analysis of real-world data
from two Portuguese water utilities. Moving forward, a weighted parallel strategy for com-
bining multiple ML algorithms is introduced to improve the prediction performance. Moreo-
ver, additional data related to weather, seasonality, and feature extraction (forecast window
of time-series data) are also analyzed. In Nasser et al. (2020), the water demand prediction is
performed using an LSTM model based on analyzing data gathered from intelligent IoT water
meters. A cloud platform has been used to store water consumption records, enabling near
real-time data streaming and storing. The performance of LSTM has been then compared to
those of SVR and RF. Similarly, in Du et al. (2021), Du et al. propose an LSTM model that
combines discrete wavelet transform (DWT) and PCA to forecast daily urban water demand.
Therefore, after smoothing the outliers of water demand time-series, noise components are
removed using DWT and pCA. Following, the LSTM network is deployed to predict urban
water demand using the outputs of DWT and PCA. Table 9 presents the main AI-Big data
analytics frameworks proposed for water management in buildings.

3.6 Evaluation metrics

Evaluation metrics are used to measure the performance of the model in terms of the
quality of its output as per what is expected. For AI applications, there are various
types of metrics that can be used based on the subject matter. That is, the outputs of
an AI model can take two forms, which are categorical variables (Cvars) and quantita-
tive variables (Qvars). For instance, the outputs of classification models represent cat-
egorical variables in which the input data are classified into different groups or classes
which are characterized by a unique label or value such as detection problems, rec-
ommender systems, etc. Each observation can be placed in a single category, and the
categories are mutually exclusive. Hence, the performance of the model depends on its
ability to correctly classify the observations to their respective categories/groups.
On the other hand, quantitative variables represent numerical values that exhibit
quantitative characteristics. Regression models have quantitative variables as outputs
such as forecasting and estimation models in which the AI model is used to represent
the mapping between the independent variable(s)—i.e., the input(s)—and the depend-
ent variable(s)—i.e., the output(s)—. In this case, the quality of the model is measured
by the closeness of the model’s outputs to the ideal expected values. Table 10 presents
a summary of the common metrics used to evaluate AI models.

13
 Table 9  A summary of AI-Big data analtiycs models for water monitoring
Ref. AI model Task Building nature Year Description Evaluation metrics Others
RMSE MAE MAPE ACC​ F1

Altunkaynak and MLP Water demand pre- Predict water demand ✓ CE

Nigussie (2017) diction using MSA-MLP
and compare its
DWT-MLP
Zubaidi et al. (2018) GSA-ANN, BSA- Water demand pre- Residential 2018 Predict water demand ✓
ANN diction using heuristis
algorithms, ANN
and weather vari-
ables
Chen et al. (2020) DT, NB, LR, WQP HEP plant 2020 Predict water ✓ wF1
LDA,CRT, KNN, quality using
SVM, RF, CRF different water
paranmetersi.e. pH,
DO, CODMn, and
NH3–N
AI‑big data analytics for building automation and management…

Shine et al. (2018) RF, NNN, SVMCDT Water consumption Agricultural 2018 Predict water con- ✓ ✓
prediction sumption using a
backward sequen-
tial variable selec-
tion and parameter
tuning
Smolak et al. (2020) RF, SVM, ARIMA Water consumption Residential 2020 Predict water con- ✓ ✓ ✓ ✓
prediction sumption using
consumption
records and occu-
pancy patterns
4985

13
 Table 9  (continued)
4986

Ref. AI model Task Building nature Year Description Evaluation metrics Others
RMSE MAE MAPE ACC​ F1

13
Antunes et al. (2018) ANN, RF, SVM Water demand pre- Public 2018 Reliable prediction ✓ ✓ R2
KNN diction while no signifi-
cant anomalies of
the data used
during training are
reported
Roccetti et al. (2019) RNN Predicting water - 2019 Predict water meter ✓ ✓ AUC, CM
meter failures failures using 15
million of readings
Nasser et al. (2020) LSTM Water demand pre- Public and residential 2020 Predict energy ✓ ✓ ✓
diction demand by analyz-
ing data gathered
from smart IoT
water meters and
stored in the cloud
Du et al. (2021) LSTM-DWT-PCA Water demand pre- Public and residential The outputs of DWT ✓ ✓ EVS, R2
diction and PCA are fed
into an LSTM
network to predict
water demand
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4987

4 Critical discussion and current challenges

A truly smart building combines a BAMS with intelligent data analytics software that
offers helpful insights for maintenance, service, and efficiency opportunities. Typically,
these tools together offer benefits for building owners, such as: (i) providing a high-
level, system-wide big data capture of the entire operations, (ii) ensuring air quality
control and a healthier building environment, (iii) saving energy and energy consump-
tion during off-peak or low occupancy periods, (iv) eliminating waste from everyday
system usage through intelligent sensor data, (v) offering guidance for performance
improvements for individual assets, (vi) addressing equipment that really needs repair
and not just those on a fixed schedule that don’t need to be serviced, and (vii) offering
advanced automation capabilities and actionable results. In addition to these benefits,
the cost savings related to the use of smart data analytics can be significant.
However, various gaps specific to each application field of AI-big data analytics
are identified. Among them, more effort should be put to efficiently carrying out text
analytics on operators’ work-order logs; and identifying (i) the information to derive,
(ii) the text-mining methods to adopt, and (iii) the efficient approaches to convey the
information to the operator and visualize it. Moreover, another challenging issue con-
cerns the use of virtual metering, where a limited number of works were dedicated to
virtual meter development (Kim et al. 2021; Wilcox 2020) despite its significance in
helping operators for understanding plant-to-zone water and energy flows and ranking
their operational decisions, such as identifying and evaluating faults.
The HVAC prognostic and failure prediction is another application that is still very
challenging, where limited research activity was conducted to target this challenge. In
fact, developing prognostics models is valuable for (i) predicting the time-to-failure,
(ii) avoiding global failures in key BAMS components (e.g. boilers, chillers, pumps,
fans, etc.), and preventing disruptions in building services. In addition, there are new
challenges from emerging BAMSs that need to be addressed, e.g. data benchmarking,
big data security and privacy, scalability and interoperability, real-time big data intel-
ligence and knowledge transfer.

4.1 Data quality issues

Usually, raw data gathered from BAMSs can have some data quality problems, including
(i) outliers, (ii) noise, (iii) inconsistent data, (iv) duplicate data, and (v) missing values.
Data pre-processing techniques are deployed to overcome these issues, such as formatting,
cleaning, and resampling. Formatting aims at converting the raw data into appropriate for-
mats to ease the application of ML algorithms, while cleaning refers to removing or replac-
ing missing samples (Zhang et al. 2021). Lastly, resampling can be applied based on the
requirements of ML algorithms. Typically, it can be (i) a down-sampling to reduce data
redundancy, foster the processing and improve the accuracy; or (ii) an up-sampling that
helps increase the amounts of data to train data-hungry ML models, especially DL algo-
rithms (Elnour et al. 2022). To that end, because of the high requirements for data quality
set by ML models, developing novel strategies to improve the quality of BAMS recorded
data by creating additional data with enhanced quality or augmenting existing datasets is a
crucial challenge.

13
 Table 10  A summary of the common evaluation metrics of AI models
4988

Metric Description Values type Application Formula

13
Relative error (RE) The absolute error between the actual Qvars Regression
and estimate values of a variable to
|(y − ŷ )∕̂y|

its estimate value

Relative change The amount of the absolute difference Qvars Regression
as a fraction of the variable’s refer-
( )
|y − yref |∕yref × 100%

ence value
| |

Mean absolute error/difference (MAE or MAD) Measures the absolute difference Qvars Regression 1
m
between predicted and actual vari-
∑m �

ables describing the same phenom-

i=1 �y
̂ i − yi ��

enon.
Mean absolute percentage error (MAPE) the MAE expressed in percentage. Qvars Regression MAE × 100%
Median Absolute Error (MedAE) Measures the median absolute error Qvars Regression
between predicted and actual values.
( )

1
Median ||ŷ i − yi ||

Mean squared error/difference (MSE or MSD) Measures the average difference Qvars Regression yi − y i )2
m i=1 (̂
between the predicted and actual
∑m

values.
Root mean squared error/difference (RMSE or The square root of the MSE to inter- Qvars Regression 1 m
m
yi
i=1 (̂ − y i )2
RMSD) pret the error in the same unit of the
� ∑

variable.
Root mean square percentage error (RMSPE) Represents the RMSE expressed in Qvars Regression RMSE × 100%
percentage
Normalized root mean squared error (NRMSE) Refers to the normalized RMSD to Qvars Regression RMSE∕(ymax − ymin )
facilitate the comparison between
variables with different scales.
R-squared (R2) Measures the fit quality of a regres- Qvars Regression 1− yi
i=1 (̂ − yi )2 ∕ yi
i=1 (̄ − y i )2
sion model/function by representing
�∑m ∑m �

the proportion of the variance for

a dependent variable in terms of
independent variable(s).
Y. Himeur et al.
 Table 10  (continued)
Metric Description Values type Application Formula

Theil U1 index Measures the relative accuracy Qvars Regression

between the actual and predicted i=1 ŷ i − yi ∕ i=1 y2i
�
∑m � �2 ∑m

results.
√ ( )
Theil U2 index Measures the quality of the predicted Qvars Regression 1 ∑m ( )2 √ 1 ∑m 2 √ 1 ∑m 2
m i=1 ŷ i − yi ∕ m i=1 yi + m i=1 ŷ i
results.
Accuracy (ACC) Measures the closeness between the Cvars Classification (TP + TN)∕(TP + TN + FP + FN)
predicted values and the targets.
Error rate (ERR) Measures the proportion of the false Cvars Classification (FP + FN)∕(TP + TN + FP + FN)
predictions in total predictions.
Precision (PPV) Measures the closeness the set of Cvars Classification TP∕(TP + FP)
predicted result
Recall or True positive rate (TPR) Measures the proportion of correct Cvars Classification TP∕(TP + FN)
positive (TP) predictions in the true
positive class.
False-positive rate (FPR) Measures the proportion offalse Cvars Classification FP∕(FP + TN)
positive (FP) predictions in the true
AI‑big data analytics for building automation and management…

negative class.
True-negative rate (TNR) Measures the proportion of correct Cvars Classification TN∕(TN + FP)
negative (TN) predictions in the true
negative class.
False-negative rate (FNR) Measures the proportion of false Cvars Classification FN∕(TP + FN)
negative (FN) predictions in the true
positive class.
F1-score Measures the harmonic average of the Cvars Classification 2(PPV × TPR)∕(PPV + TPR)
precision and recall
Matthews correlation coefficient (MCC) Measures the quality of binary clas- Cvars Classification TP×TN−FP×FN
(TP+FP)(TP+FN)(TN+FP)(TN+FN)
sifications.
√
4989

13
 Table 10  (continued)
4990

Metric Description Values type Application Formula

1

13
Prediction interval coverage probability (PICP) Evaluates whether the actual value Qvars Regression
N i=1,
𝛼 is within the prediction interval
∑N

limits,𝛼i = 1 if it lies within the pre-

diction interval, 𝛼i = 0 otherwise
Prediction interval normalized average width Measures the width of the prediction Qvars Regression 1
NE i=1 Ui − Li
(PINAW) interval with Li and Ui being the
∑N � �

lower and the upper boundaries,

respectively, and E is the difference
between the maximum and the mini-
mum actual values.
Precision recall curve (PRC) Used to present the trade-off between Cvars Classification –
precision and recall using different
thresholds.
Receiver operating characteristic curve (ROC) Used to present the trade-off between Cvars Classification –
the false positive rate and true posi-
tive rate using different thresholds.
Area under the ROC (AUROC) Represents the area under the ROC Cvars Classification –
curve and the greater the value, the
better the classification performance
Cross validation (CV) A resampling procedure to evaluate the Cvars and Classification and –
generalization ability of an AI model Qvars Regression
Confusion matrix A tabulated representation of clas- Cvars Classification –
sification of results of the algorithm
under study.
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4991

4.2 Data scarcity and data benchmarking

The different applications of BAMSs necessitate extensive historical data to train the
AI-big data analytics, especially those based on DL algorithms before they can be used
reliably. However, large-scale data might not be available for some reason or can not be
recorded representatively and sufficiently in a short time when we study newly-built envi-
ronments. Fortunately, the problems addressed within each specific AI-big data analyt-
ics task in BAMSs illustrate some similarities. This could be justified by the fact that the
different application tasks, despite studying distinct problems, use the same data-driven
algorithms which are validated on slightly similar datasets collected from different kinds
of buildings and devices. This has opened opportunities for using knowledge transfer and
transfer learning to overcome the lack of datasets in some situations, e.g. sports facilities.
On the other hand, collecting and benchmarking data represents the most significant
challenge so far when applying AI-big data analytics in BAMSs, especially for the case of
large buildings, i.e. sports facilities, commercial centers, industrial buildings. Many tasks
require annotated datasets to train AI models and validate them. Indeed, developing and
validating new data-driven algorithms require recording and annotating large-scale data-
sets, especially when using DL algorithms that are notoriously data-hungry (Kučera and
Pitner 2018). Improving the performance of BAMSs does not rely only on the selection of
AI algorithms but also on the quality and parameters of datasets used to train them.
For instance, labeled and accurate anomaly detection datasets are needed for develop-
ing new automatic anomaly detection solutions. Similarly, development and validation of
occupancy detection algorithms require repositories of building occupancy profiles with
concurrent ground-truth people counts. In this context, to further improve the performance
of BAMSs, public and open-access datasets are needed for different application fields (e.g.,
load forecasting, anomaly and fault detection, demand response, occupant-centric controls,
IEQ monitoring, water monitoring, etc.) for assessing the AI-big data analytics algorithms
developed by the AI Community (Park et al. 2019; Francisco et al. 2020).
From another hand, successful sustainability strategies to overcome this issue could be
via (i) incentivizing buildings/facilities managers for participating in benchmarking cam-
paigns and surveys organized for different application fields, and (ii) encouraging the AI
community in organizing data benchmarking competitions and challenges.

4.3 Security and privacy preservation

The nature of data collected in BAMSs introduces new challenges in data analytics, i.e.
security and privacy preservation, in which traditional technologies can not deal with.
Using encapsulated protocols and IP-based communication, BAMSs are more and more
connected to corporate networks and also remotely accessed for management reasons, both
for emergency and convenience purposes. However, security and privacy preservation have
not been set as a primary concern when designing these protocols. Therefore, most of the
BAMSs are being operated with sub-standard or non-existent security implementations,
and mainly rely on ensuring security by obscurity. In this line, there has been recently a
move to address the shortfalls of security and privacy preserving implementations in
BAMSs (Stamatescu et al. 2020; Ashaj and Erçelebi 2020). However, the definition of the
new threats against BAMSs, and identification of these threats is still a field that is excep-
tionally lacking.

13
4992 Y. Himeur et al.

Moreover, another critical concern about security in BAMSs is related to the fact that
buildings’ data is valuable not only for managers and other BAMS competitors, as it is
attached to the control of buildings’ equipment. Typically, it could be significantly criti-
cal to end-users if manipulated. To that end, sharing data in most BAMSs has limited the
buildings’ intranet. Also, any attempts to extract this data to cloud data centers can result in
severe security risks, considerably higher costs for the appropriate security systems or both
(Lv et al. 2021; Himeur et al. 2022).

4.4 Scalability and interoperability

An important issue of BAMSs is the inherent lack of scalability and interoperability. This
is because each BAMS manufacturer has its own proprietary data protocol that requires
the development and maintenance of various processes and integrations. Moreover, BAMS
vendors usually have competing products and thus are incentivized to make their data inac-
cessible to third parties (Png et al. 2019; Tang et al. 2020). Therefore, the interoperabil-
ity is a legitimate concern for making efficient smart buildings as it refers to the ability
of all the systems inside a building to communicate with one another. Specifically, with
the actual proliferation of intelligent building technologies designed and manufactured by
a plethora of companies, there is a need to make them communicate universally for pro-
moting their deployment inside residential, commercial, industrial, and office buildings
(Ozturk 2020; Miori et al. 2019). Put differently, as smart buildings need to meet energy
and water efficiency, adequate indoor environmental conditions, high comfort levels, and
economic goals set by building managers/users, they require the use of a highly-connected
building automation system in which different parts can efficiently communicate with one
another and adjust to changes in the environment. However, while numerous buildings are
equipped with excellent systems, they often lack a combined monitoring system for light-
ing, climate control, water monitoring and blinds that could facilitate efficiency measures
(Schachinger et al. 2017).
The BAMS community makes great efforts to develop and deploy communication pro-
tocols that ensure interoperability, such as Modbus, KNX, LonWorks, and Protocol 3964R
(Merz et al. 2018). For instance, LonWorks and KNX are interoperable open standards as
they can be used together. However, integration concerns may arise (Tang et al. 2020). The
technologies’ popularity can hinder interoperability potential in BAMSs, as the example
with LonWorks, the market leader in the United States. In contrast, KNX—widely used in
Europe—has yet to impact (Merz et al. 2018). Additionally, the installation and operation
costs can pose severe integration limitations. Indeed, many contractors or system integra-
tors that provide off-the-shelf solutions are less concerned about whether the integration is
successful.

4.5 Real‑time big data intelligence

Collecting and analyzing data in real-time are of significant importance while designing
powerful and efficient BAMSs. The first step towards this is by adopting a real-time sub-
metering, which helps in tracking track utility costs by building region (floor, room, etc),
by the tenants, by individual facility equipment (e.g., HVAC, lighting), etc. Therefore,
granular utility sub-metering data provides the essential tools for monitoring energy costs/
performance, water consumption/waste. This will result in accurately identifying usage

13
AI‑big data analytics for building automation and management… 4993

anomalies, enabling data-driven portfolio analysis, etc. Although the value of real-time
sub-metering is unquestioned, most of the BAMSs are still unable to provide the real-time
data monitoring, which delays decision-making measures and hence reduces then the qual-
ity of efficiency and optimization operations.

5 Case studies

This literature review established several applications of AI big data analytics for build-
ings in terms of energy management, load forecasting, water management, FDAD, or IEQ
monitoring. In this section, we present their deployment for energy-related applications
given the continuous rise in energy consumption worldwide of the buildings sector under
the global energy dilemma and the energy optimization potential of BAMSs, given the
increasing concerns about energy efficiency in buildings. More specifically, the case stud-
ies handle two of the lead causes of energy waste in buildings, which are (i) system faults
and equipment malfunctioning, and (ii) poor management and regulation of the buildings’
systems (Alsalemi et al. 2022; Elnour et al. 2022, 2020). The first two case studies pre-
sent strategies for energy anomaly detection that can be due to both or either of the for-
mer causes. They demonstrate the deployment of two different methods: unsupervised and
supervised learning. The last case study presents the use of AI data analytics to establish
reliable and efficient regulation of HVAC systems, given that those are considered major
energy consumers in buildings (energy.gov, Elnour et al. 2022; Fadli et al. 2021).

5.1 Unsupervised AI‑based energy anomaly detection

This section presents an example of using unsupervised ML algorithms for detecting

abnormal energy consumption (Himeur et al. 2022). Therefore, four algorithms are con-
sidered, namely (i) OCSVM with linear kernel, (ii) OCSVM with Gaussian kernel, (iii)
DBSCAN, and (iv) LOF. They have been applied on the Dutch residential energy data-
set (DRED), which incorporates electricity consumption, occupancy patterns and ambient
conditions of a typical household (in the Netherlands). Figure 6 portrays the scatter plot
of energy footprints in which normal and abnormal patterns are identified using the afore-
mentioned approaches. It has been clearly seen OCSVM (with linear kernel) detects more
energy samples that fall outside the inlier region, which refer to consumption anomalies.
While by using OCSVM (with a Gaussian kernel), the number of samples that fall inside
the inlier region has been reduced because of its separation capability introduced by the
hyperplane generated using the Gaussian kernel. From another side, LOF and DBSCAN
help detect abnormal patterns with almost the same efficiency as OCSMV (with the Gauss-
ian kernel), and only a slight difference has been registered in classifying a few numbers of
samples.

5.2 Supervised AI‑based energy anomaly detection

Supervised ML algorithms excelled in detecting abnormal energy usage, although they

require labeled energy data. To that end, in Himeur et al. (2021), a micro-moment-
based approach is introduced to cluster energy footprints of an office building (at Qatar
University) into five classes with reference to the energy consumption, occupancy pat-
terns and appliance operation specifications. These classes are named ”class 0: good

13
4994 Y. Himeur et al.

Fig. 6  Energy consumption anomaly detection in residential buildings using a) OCSVM with linear kernel,
b) OCSVM with Gaussian kernel, c) DBSCAN and d) LOF

usage”, ”class 1: turn on appliance”, ”class 2: turn off appliance”, ”class 3: exces-
sive consumption” and ”class 4: consumption while outside”. Following, an improved
KNN model is developed and used to learn abnormal energy usage using this anno-
tated data. Fig. 7 illustrates the flowchart of the micro-moment based scheme used
to extract and learn intent-driven moments of energy consumption. Typically, energy
micro-moment features MF are extracted based on analyzing occupancy profiles (O)
and power consumption (p) of each device in reference to device active consumption
range (DACR​), device operation time (DOT) and device standby power consumption
(DSPC). Then, the appliance operation parameters are called, including DACR​, DOT
and DSPC. Table 11 presents an example of different appliance parameter specifica-
tions that are used in the rule-based algorithm to extract power consumption micro-
moments (Himeur et al. 2022).
To have a clear view of how abnormal energy consumption is distributed over
the time, the scatter plot of energy consumption profiles of a television is illustrated
in Fig. 8. Accordingly, the corresponding normal and abnormal energy patterns are
detected using an IKNN model and micro-moment analysis. Because this approach
uses a supervised learning with regard to occupancy data, it has the capability of

13
AI‑big data analytics for building automation and management… 4995

Fig. 7  Block diagram of the supervised ML solution used to detect abnormal energy consumption in office
buildings

identifying new consumption anomalies that correspond to the absence of the end-
users when the television is on (this abnormality can be extended to other devices that
require the presence of the user during their operation, e.g. the air conditioner, heater,
fan, etc.). Detecting such abnormalities was not doable if an unsupervised ML model
was deployed, in which only energy patterns were analyzed.

5.3 Energy and performance optimization for sports facilities

In light of the increased global energy demand and its associated environmental impacts,
the management and optimization of sports facilities are becoming imperative as they are
characterized by high energy demand and occupancy profiles. This case study demonstrates
the application of the model predictive control (MPC) theory and NNs for energy and per-
formance management of sports facilities. Figure 9 presents the proposed NN-based MPC
framework. The work is carried out using the building information model of a sports hall
in the sports complex of Qatar University using EnergyPlus and practical data for model
calibration. MPC systems are robust as they allow integrated dynamic optimization that
accounts for the future system behavior in the decision-making process. NNs are advanta-
geous for their ability to represent complex functions with high accuracy.

13
 4996

13
Table 11  Power consumption specifications for different home appliances
Appliance DOT DACR (watts) DSPC (watts) Appliance DOT DACR (watts) DSPC (watts)

Air conditionner 15 h 30 min 1000 4 Washing machine 1h 500 6

Microwave 1h 1200 7 Light 8h 60 0
Oven 3h 2400 6 Television 12 h 42 min 65 6
Dishwasher 1h 45 min 1800 3 Refrigerator 17 h 30 min 180 0
Laptop 12 h 42 min 100 20 Desktop 12 h 42 min 250 12
Y. Himeur et al.
AI‑big data analytics for building automation and management… 4997

Fig. 8  Scatter plot of energy micro-moments identified using IKNN (Himeur et al. 2021)

Fig. 9  Block diagram of the NN-based MPC framework for sports facilities energy and performance opti-
mization

13
4998 Y. Himeur et al.

(a) (b)

(c)

Fig. 10  The performance of the NN-based MPC system for energy and performance optimization in the
sports hall in Qatar University sports complex

The NN-based dynamic prediction model aims to express and capture the behavior of
the building operation over time given its states x(k) (i.e., power usage, thermal comfort,
indoor and outdoor air properties, etc.) and its inputs (i.e., HVAC system settings). The
NN-based prediction model is:
F = Train_NN(x(k + 1), [x(k), u(k + 1)]), (1)
and the prediction is computed by:
x̂ (k + 1) = F(x(k), u(k + 1)). (2)
The optimization of the NN’s hyper-parameters was performed using the Bayesian optimi-
zation algorithm which keeps track of past iterations to find better choices for the next set
of hyper-parameters to evaluate (Andonie 2019).
The MPC system consists of an optimizer and an NN-based prediction model of the
building operation, and based on the system output y(k) ⊂ x(k) and its reference value
r(k) (i.e., power usage and thermal comfort level), the HVAC system settings for tem-
perature setpoints and dampers positions are determined (i.e., u(k)) using numerical
optimization to achieve tracking. When compared to routine performance, the proposed
approach was able to achieve significant energy reduction and adequate thermal comfort
levels as demonstrated in Fig. 10. Energy savings of around 15% was observed, which
was approximated by evaluating the relative change in the total energy consumption in
the two settings for the scenario under study, that is, the relative difference between
the areas under the two power curves in Fig. 10a. Considerations about the NN model

13
AI‑big data analytics for building automation and management… 4999

Fig. 11  Block diagram of the improved CUSP model with efficiency-comfort-health model

Fig. 12  The web interface of the expanded and improved CUSP platform with the three integrated models
for efficiency, comfort, and health & safety

performance, tuning of the MPC settings, and optimization sub-optimality or failure are
essential during the design and implementation phases of the proposed framework.

5.3.1 Improved computational sustainability model for sports facility management

The computational urban sustainability platform (CUSP), developed at Cardiff University,

is an immersive decision support tool built to deliver a powerful urban analytics and enable
interactive monitoring and inform decision making through a web interface. It can be used
to promote co-simulation across disciplines, and predict future scenarios towards a sustain-
able future operation and urban Intelligence [Computational Urban Sustainability Platform
(CUSP)]. The CUSP model can be improved to include three integrated models, which are

13
5000 Y. Himeur et al.

(1) energy-water efficiency, (2) health, safety, and wellbeing, and (3) comfort as demon-
strated in Fig. 11.
The improved CUSP model integrates an energy simulation tool that is used to generate
data of the particular scenario under consideration for data analytics for quality monitoring
and planning purposes. It contains three AI-based models developed to assess each of the
three aspects of the facility operation, which are efficiency and sustainability, health and
safety, and users’ thermal satisfaction. Through the web interface shown in Fig. 12, the
integrated simulation tools will enable facility managers to evaluate the possible scenario
in terms of the HVAC system settings, and occupancy and operation schedules towards
achieving a reasonable trade-off between those three aspects prior to applying them in the
facility.

6 Future directions

6.1 Multimodal data analysis

Due to the advancement of today’s sensing and mobile technologies, various modalities
of data can be easily and effectively gathered using different and advanced means. Thus,
it is now possible to record and process big data about environmental satisfaction lev-
els of buildings’ occupants in real-time and non-invasive manners (Plageras et al. 2018).
Buildings’ end-users naturally react to ambient environmental conditions for minimizing
any environmental stress, increasing their comfort based on their autonomic nervous sys-
tems and expressed by different poses, which can effectively influence different building
operation parameters (Amato et al. 2018). Therefore, it becomes of utmost importance to
develop tools for enhancing the interdisciplinary knowledge (i.e. AI, IoT, big data, DL,
computer vision) when managing building operations. This helps significantly advance
building indoor environmental control and sensing technologies as a function of human
bio-signals (i.e. physiological signals) and poses.
Analyzing multi-modal data helps BAMSs in boosting workplace productivity and opti-
mising office spaces, which in turn cutting costs and increasing revenues for companies.
Moreover, data generated from these systems could be used to reduce the spread of viruses
and other diseases inside buildings, increasingly important since the outbreak of Covid-19
(Sun and Zhai 2020). For instance, in Ding et al. (2020), investigate the collective con-
tagion of the COVID-19 virus inside indoor environments (i.e. healthcare facilities and
public vehicles) along with the engineering control against virus spread with ventilation
systems.

6.2 In‑situ sensor calibration in BAMSs

Sensors are key players in helping BAMSs to attain expected efficiency and automation.
However, they are affected by continuous failures and degradation over time. To that
end, in-situ sensor calibration plays a crucial role in calibrating different BAMS work-
ing sensors (i.e., physical sensors) and avoiding significant errors for reliable results
when it is deployed to large-scale building sensor networks (Yu and Li 2015). Most
of the studies opted for the conventional periodical calibration as a solution to over-
come sensor degradation and failure; however, this is impractical and difficult for vari-
ous sensors. By contrast, virtual in-situ calibration (VIC) can be a good alternative since

13
AI‑big data analytics for building automation and management… 5001

it relies on mathematically extracting the characteristics of essential aspects involved

in a calibration, such as the uncertainty quantification, benchmark establishment, and
environment assessment (Yoon and Yu 2018). Moreover, because BAMSs need digitally
enhanced data-rich environments, virtual sensors offer reliable and informative sens-
ing contexts for operational datasets in BAMSs. More specifically, in-situ virtual sen-
sors help develop the counterparts of target physical sensors in the field. Therefore, they
can provide extra data related to residuals between physical and virtual sensors for for
deployment in data-driven modeling, diagnostics and analytics (Koo et al. 2022).

6.3 Smart building digital twins

The increasing amounts of data generated by BMAMs, and the need for new methods
to leverage it, have motivated scientists to investigate new strategies. One promising
solution is using the digital twins (DT) paradigm, which assumes a complete cohesion
and integration between the visual and physical worlds. Typically, DT can deliver con-
siderable benefits to the BAMSs and the built environment in general by helping bring
together static and dynamic data from various sources (in 2D/3D models) and assisting
in making effective and informed decisions. Moreover, it combines the knowledge from
the physical and digital worlds by collecting real-time data from the physical environ-
ments and provides a real-time understanding of buildings’ performance (Delgado and
Oyedele 2021). Besides, despite the gradual exploration of digital twinning within the
fields of building information modeling (BIM) and cyber-physical systems (CPS), avail-
able tools and techniques need to be considered in the next level of integration (technol-
ogies and procedures). This is to (i) provide DTs with more adaptability and more cohe-
sion over the managed information and (ii) extract more value from our virtual models
(Shahzad et al. 2022).

6.4 Transfer learning

Specifically, transfer learning has recently been proposed as solution that can be investi-
gated for the case of buildings with poor information data (Himeur et al. 2022). Put sim-
ply, data and knowledge of already existing buildings (or old buildings) with rich energy
usage records, water management data, occupancy patterns, IEQ monitoring footprints
and and ambient environmental conditions can be used. Therefore, various frameworks
have been introduced for target energy forecasting (Gao et al. 2020; Li et al. 2021),
anomaly detection of energy consumption (Liang et al. 2018; Xu et al. 2021), fault diag-
nosis of energy systems (Liu et al. 2021; Zhu et al. 2021), HVAC fault detection (Dowl-
ing et al. 2020), IEQ monitoring (Tariq et al. 2021), indoor occupancy detection (Khalil
et al. 2021), etc.

6.5 Blockchain

Due to the security and privacy issues that are still open in BAMSs, blockchain is consid-
ered as a promising solution that provides the digital trust. It can function as a permanent,

13
5002 Y. Himeur et al.

cloud-based and digital ledger of activities between different users and partners (Nawari
and Ravindran 2019; Liu et al. 2021). Also, blockchain can operate as a distributed, single
source of shared truth and has the possibility of becoming the top-system for recording
all transactions. Therefore, its deployment in BAMSs aims at (i) tracking and validating
changes (e.g. security and surveillance, access control, etc.), (ii) monitoring HVAC activi-
ties, (iii) recording property transfers, and (iv) detecting occupancy patterns (Siountri et al.
2020). Additionally, it can help manage intelligent buildings and IoT devices with renew-
able energy, e.g. wind and solar. For example, suppose a facility is in a two-way energy
communication with the grid. In that case, blockchain can make it more secure and easier
to develop a digital record of energy-in and energy-out transactions (Tiwari and Batra). On
another side, as the global market of building automation exceeds $120 billion, smart con-
tracts can be utilized for automating warranties and providing refunds when IoT-connected
devices or equipment do not perform as expected (Himeur et al. 2022).
Overall, there are numerous potential applications of blockchain in BAMSs, although
the principal advantages are data is easy to access, is secure, and can not be corrupted.
Specifically, data stored in the blockchain database can be easily and quickly reviewed,
even though it is managed by distinct entities, which results in an accurate and fast data
analysis (Nawari and Ravindran 2019). Moving forward, blockchain helps in stream-
lining processes and lowering costs through reducing and/or eliminating those dreaded
manual operations, especially in public buildings, sports facilities, and commercial
centers. This could be adapted to almost any process, including preventive maintenance,
work orders, environmental health, and safety planning, and space management (Nawari
and Ravindran 2019).
Only for energy management in smart buildings, blockchain has found diverse applica-
tions. For instance, in Van Cutsem et al. (2020), use a blockchain-based approach to coop-
erate energy management of multiple end-users in smart-buildings, where smart-contracts
have been utilized to allow decentralizing community energy management. In Mukherjee
et al. (2021), a smart energy management solution is safeguarded with blockchain and
hence ensures judicious generation, uniform distribution and shielded monitoring along
with guaranteed security and privacy of the havoc data. In Tiwari and Batra, blockchain is
introduced for enabling the reparation of smart buildings-cyber physical systems. Moving
on, decentralized and flexible access control using smart contracts is developed for smart
and large commercial buildings in Bindra et al. (2021). This solution has been proposed
as an alternative to inefficient, unsystematic, and human-intensive access control schemes
usually used in these buildings. While the widespread implementation of blockchain is
still a long way off, it is also challenging to deploy this technology reliably and widely in
BAMSs. This research area needs to be further investigated in the near future. This prom-
ising new technology could benefit the other tasks of smart buildings, i.e. water manage-
ment, IEQ monitoring, occupancy detection, etc.

6.6 Cyber‑security standards for BAMSs

While using AI in BAMSs represents a powerful asset, it also presents some data secu-
rity and privacy concerns and problems with the regulations. Typically, AI-driven
BAMSs involve the deployment of lower-cost sensors (both wired and wireless) and the
adoption of cloud, fog, edge, and/or hybrid computing architectures, increasing cyber
risks. To that end, the need for a sound cybersecurity strategy has become crucial for
promoting secure remote BAMSs. Data flows must be planned and monitored, possibly

13
AI‑big data analytics for building automation and management… 5003

making it necessary to use one-way data diodes. On the other hand, BAMSs integrate
heterogeneous sensing, computation, and control capabilities. They combine cyberspace
with the physical world to develop cyber-physical systems. However, the security of
BAMSs is significantly threatened by software/hardware failures and/or cyber/physical
attacks. For example, sensor failures can engender false detection of abnormal energy/
water consumption behaviors and result in actuator misbehavior.
To handle the above issues, privacy and security protection mechanisms should be
enforced. This is possible by providing recommendations to the building automation
community, e.g., the data protection directive 95/46/EC (Tokarski) suggests recommen-
dations for supporting the security of the implementation of smart metering and smart
using data controllers. Addressing these recommendations can enable moving to fully
harmonized data protection environments and improving security measures in BAMSs.
Moreover, different cyber security standards can be used to secure BAMSs by address-
ing the cybersecurity for operational technology in automation and control systems,
such as ISA/IEC 62443 series (Bicaku et al.) and ASHRAE 135 series (BACnet) (Tang
et al. 2020).

6.7 Self‑learning for long‑term building operation

Self-learning ML models are key to realizing the BAMS in the long-term building
operation. The systems built upon these models have recently gained industry recogni-
tion and market share as they are based on using a ”user-friendly” technology (Cortiços
2019). Typically, self-learning, also called self-supervision, is an emerging technology
that helps develop computationally efficient, low-cost, autonomous, and self-supervised
ML algorithms (Kaklauskas et al. 2019). For example, for energy management, a self-
learning control scheme assists in assessing the energy flexibility of buildings, in addi-
tion to guaranteeing robustness, scalability, and adaptability. Moreover, automated
self-learning systems have promising perspectives when they are to integrate demand-
response strategies for effective home-energy management systems (Bampoulas et al.
2019).

6.8 Edge analytics for BAMSs

With the advancement of BAMSs and the latest generation of IoT devices, data acquisi-
tion from multiple types of equipment has become much easier in today’s buildings.
Real-time access to this data helps in better managing facility operations, sustaining
efficiency, and lowering costs. However, as most BAMSs are only implemented using
cloud computing, real-time data analysis may not be guaranteed. To overcome this issue,
open software platforms hosted on edge nodes in close proximity to the building’s IoT
devices can enable access to the building data and advanced analytics deployed on these
platforms in real-time. In this context, edge computing employs the processing power
of IoT devices for filtering, pre-processing, aggregating and storing recorded data, and
actions can then be performed in real-time using adequate analytical algorithms. This is
because edge computing enables resolving bandwidth and latency problems and reduc-
ing response time. Following, the filtered data could be transmitted to the cloudlet plat-
forms for aggregation and enrichment, and running of complex analytics (Sharma et al.
2018).

13
5004 Y. Himeur et al.

Thus, various use cases where edge computing and the IoT can efficiently be utilized in
BAMSs are emerging, among them fault diagnosis, which helps to (i) find patterns in sen-
sor data representing equipment failures, anomalies, or degraded performance; (ii) detect
abnormal energy consumption, e.g. if the lighting or HVAC systems are activated too early
or operate too late with regard to the actual occupancy schedules; and (iii) identify correla-
tions across different types of data, which are essential to infer the factors impacting energy
consumption (e.g. the patterns related to weather, age of facilities, etc). Overall, open edge
software platforms combine multi-protocol connectivity and the ability to aggregate data
from multiple sources and facilitate the task of advanced analytics in turning this data into
actionable information that can be used to improve the overall operational efficiency build-
ings (Petri et al.).

7 Conclusion

This paper carried out a comprehensive overview of the application of AI-big data analyt-
ics in BAMSs to conduct different tasks, including energy forecasting, fault and anomaly
detection, water monitoring, and IEQ monitoring. The pros and cons of AI models within
the unsupervised, supervised, semi-supervised, and reinforcement learning categories have
been identified. Moreover, it concluded that supervised learning algorithms excelled well
in performing the diver BAMS tasks, but their performance always relies on the availabil-
ity of annotated data and its accuracy. Unsupervised learning models with no prior knowl-
edge can address this issue with less efficiency.
It was demonstrated in this framework that technologies of ML, IoT, and new connec-
tivity capabilities have a critical role in shaping the future of BAMSs. With building own-
ers and facility managers focusing heavily on improving energy efficiency and increasing
cost savings, features like advanced fault detection and diagnostics, energy analytics, IEQ
monitoring, and water management are becoming critical. The growing interest devoted to
developing intelligent analytics in BAMSs has been highlighted by the increasing number
of works and studies proposed in the literature to address several challenges. In the coming
years, data analytics is expected to expand the capabilities of intelligent building technolo-
gies, spurring further advancements in building automation systems and equipment stand-
ards while minimizing the environmental impact of commercial buildings.
The AI-big data analytics technology is up-and-coming to BAMSs. However, it faces
various challenges for achieving market penetration, including legal, regulatory, security
and privacy preservation, interoperability and scalability, and competition barriers. Addi-
tional research initiatives, investigations, projects, and collaborations should be considered
a primary requirement for showing if the technology can reach its absolute power, prove its
commercial viability, and lastly, be adopted in the mainstream.
Acknowledgements This publication was made possible by NPRP Grant No. NPRP12S-0222-190128 from
the Qatar National Research Fund (a member of Qatar Foundation). The findings achieved herein are solely
the responsibility of the authors.

Funding Open Access funding provided by the Qatar National Library.

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License,
which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as
you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons

13
AI‑big data analytics for building automation and management… 5005

licence, and indicate if changes were made. The images or other third party material in this article are
included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the mate-
rial. If material is not included in the article’s Creative Commons licence and your intended use is not per-
mitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http://​creat​iveco​mmons.​org/​licen​ses/​by/4.​0/.

References
Aazam M, Zeadally S, Harras KA (2018) Deploying fog computing in industrial internet of things and
industry 4.0. IEEE Trans Ind Inform 14(10):4674–4682
Abba S, Pham QB, Usman A, Linh NTT, Aliyu D, Nguyen Q, Bach Q-V (2020) Emerging evolutionary
algorithm integrated with kernel principal component analysis for modeling the performance of a
water treatment plant. J Water Process Eng 33:101081
Abdulhammed R, Musafer H, Alessa A, Faezipour M, Abuzneid A (2019) Features dimensionality reduc-
tion approaches for machine learning based network intrusion detection. Electronics 8(3):322
Aboelwafa MM, Seddik KG, Eldefrawy MH, Gadallah Y, Gidlund M (2020) A machine-learning-
based technique for false data injection attacks detection in industrial IoT. IEEE Internet Things J
7(9):8462–8471
Acquaah Y, Steele JB, Gokaraju B, Tesiero R, Monty GH, Occupancy detection for smart HVAC efficiency
in building energy: a deep learning neural network framework using thermal imagery. In: 2020 IEEE
applied imagery pattern recognition workshop (AIPR). IEEE, pp 1–6
Afaifia M, Djiar KA, Bich-Ngoc N, Teller J (2021) An energy consumption model for the Algerian residen-
tial building’s stock, based on a triangular approach: geographic information system (gis), regression
analysis and hierarchical cluster analysis. Sustain Cities Soc 74:103191
Agha G, Palmskog K (2018) A survey of statistical model checking. ACM Trans Model Comput Simul
(TOMACS) 28(1):1–39
Aghemo C, Blaso L, Pellegrino A (2014) Building automation and control systems: a case study to evalu-
ate the energy and environmental performances of a lighting control system in offices. Autom Constr
43:10–22
Ahmad T, Chen H (2018) Short and medium-term forecasting of cooling and heating load demand in build-
ing environment with data-mining based approaches. Energy Build 166:460–476
Ahmad T, Chen H (2018) Utility companies strategy for short-term energy demand forecasting using
machine learning based models. Sustain Cities Soc 39:401–417
Ahmad T, Chen H, Huang R, Yabin G, Wang J, Shair J, Akram HMA, Mohsan SAH, Kazim M (2018)
Supervised based machine learning models for short, medium and long-term energy prediction in
distinct building environment. Energy 158:17–32
Ahmad T, Huanxin C, Zhang D, Zhang H (2020) Smart energy forecasting strategy with four machine learn-
ing models for climate-sensitive and non-climate sensitive conditions. Energy 198:117283
Ahn J, Shin D, Kim K, Yang J (2017) Indoor air quality analysis using deep learning with sensor data. Sen-
sors 17(11):2476
Akil M, Tittelein P, Defer D, Suard F (2019) Statistical indicator for the detection of anomalies in gas,
electricity and water consumption: application of smart monitoring for educational buildings. Energy
Build 199:512–522
Al Dakheel J, Del Pero C, Aste N, Leonforte F (2020) Smart buildings features and key performance indica-
tors: a review. Sustain Cities Soc 61:102328
Al-Ali A-R, Zualkernan IA, Rashid M, Gupta R, AliKarar M (2017) A smart home energy management sys-
tem using IoT and big data analytics approach. IEEE Trans Consum Electron 63(4):426–434
Alawadi S, Mera D, Fernández-Delgado M, Alkhabbas F, Olsson CM, Davidsson P (2022) A comparison
of machine learning algorithms for forecasting indoor temperature in smart buildings. Energy Syst
13:1–17
Alghamdi A, Hu G, Haider H, Hewage K, Sadiq R (2020) Benchmarking of water, energy, and carbon flows
in academic buildings: a fuzzy clustering approach. Sustainability 12(11):4422
Alhussein M, Aurangzeb K, Haider SI (2020) Hybrid CNN-LSTM model for short-term individual house-
hold load forecasting. IEEE Access 8:180544–180557
Aliberti A, Bottaccioli L, Macii E, Di Cataldo S, Acquaviva A, Patti E (2019) A non-linear autoregressive
model for indoor air-temperature predictions in smart buildings. Electronics 8(9):979

13
5006 Y. Himeur et al.

Aljabre SH (2002) Hospital generated waste: a plan for its proper management. J Family Commun Med
9(2):61
Al-Kababji A, Alsalemi A, Himeur Y, Fernandez R, Bensaali F, Amira A, Fetais N (2022) Interactive visual
study for residential energy consumption data. J Clean Prod 366:132841
Alsalemi A, Himeur Y, Bensaali F, Amira A, Sardianos C, Varlamis I, Dimitrakopoulos G (2020) Achiev-
ing domestic energy efficiency using micro-moments and intelligent recommendations. IEEE Access
8:15047–15055
Alsalemi A, Himeur Y, Bensaali F, Amira A, Sardianos C, Chronis C, Varlamis I, Dimitrakopoulos G (2020)
A micro-moment system for domestic energy efficiency analysis. IEEE Syst J 15(1):1256–1263
Alsalemi A, Himeur Y, Bensaali F, Amira A (2022) An innovative edge-based internet of energy solution
for promoting energy saving in buildings. Sustain Cities Soc 78:103571
Alsalemi A, Al-Kababji A, Himeur Y, Bensaali F, Amira A (2020) Cloud energy micro-moment data clas-
sification: a platform study. In: 2020 IEEE/ACM 13th international conference on Utility and Cloud
Computing (UCC). IEEE, pp 420–425
Alsalemi A, Himeur Y, Bensaali F, A Amira (2021) Smart sensing and end-user behavioral change in resi-
dential buildings: an edge internet of energy perspective. IEEE Sensors J 21(24), 27623–27631
Altunkaynak A, Nigussie TA (2017) Monthly water consumption prediction using season algorithm and
wavelet transform-based models. J Water Resour Plan Manag 143(6):04017011
Amato G, Barsocchi P, Falchi F, Ferro E, Gennaro C, Leone GR, Moroni D, Salvetti O, Vairo C (2018)
Towards multimodal surveillance for smart building security. In: Multidisciplinary digital publishing
institute proceedings, vol 2, p 95
Andonie R (2019) Hyperparameter optimization in learning systems. J Membr Comput 1(4):279–291.
https://​doi.​org/​10.​1007/​s41965-​019-​00023-0
Antunes A, Andrade-Campos A, Sardinha-Lourenço A, Oliveira M (2018) Short-term water demand fore-
casting using machine learning techniques. J Hydroinf 20(6):1343–1366
Aparicio-Ruiz P, Barbadilla-Martín E, Guadix J, Cortés P (2021) Knn and adaptive comfort applied in deci-
sion making for HVAC systems. Ann Oper Res 303(1):217–231
Aquino I, Nawari NO (2015) Sustainable design strategies for sport stadia. Suburban Sustain 3(1):3
Ashaj SJ, Erçelebi E (2020) Energy saving data aggregation algorithms in building automation for health
and security monitoring and privacy in medical internet of things. J Med Imaging Health Inform
10(1):204–210
Aste N, Manfren M, Marenzi G (2017) Building automation and control systems and performance optimiza-
tion: a framework for analysis. Renew Sustain Energy Rev 75:313–330
Azuatalam D, Lee W-L, de Nijs F, Liebman A (2020) Reinforcement learning for whole-building HVAC
control and demand response. Energy AI 2:100020
Bampoulas A, Saffari M, Pallonetto F, Mangina E, Finn DP (2019) Self-learning control algorithms for
energy systems integration in the residential building sector, In: (2019) IEEE 5th world forum on
internet of things (WF-IoT). IEEE 2019, pp 815–818
Barrett E, Linder S (2015) Autonomous HVAC control, a reinforcement learning approach. In: Joint Euro-
pean conference on machine learning and knowledge discovery in databases. Springer, pp 3–19
Bashari M, Rahimi-Kian A (2020) Forecasting electric load by aggregating meteorological and history-
based deep learning modules. In: IEEE power & energy society general meeting (PESGM). IEEE,
pp 1–5
Bassamzadeh N, Ghanem R (2017) Multiscale stochastic prediction of electricity demand in smart grids
using Bayesian networks. Appl Energy 193:369–380
Berger MA, Mathew PA, Walter T Big data analytics in the building industry. ASHRAE J
58 (LBNL-1005983)
Bertone E, Sahin O, Stewart RA, Zou PX, Alam M, Hampson K, Blair E (2018) Role of financial mecha-
nisms for accelerating the rate of water and energy efficiency retrofits in Australian public buildings:
hybrid Bayesian network and system dynamics modelling approach. Appl Energy 210:409–419
Bessani M, Massignan JA, Santos TM, London JB Jr, Maciel CD (2020) Multiple households very short-
term load forecasting using Bayesian networks. Electric Power Syst Res 189:106733
Bicaku A, Zsilak M, Theiler P, Tauber M, Delsing J (2022) Security standard compliance verification in
system of systems. IEEE Syst J 16(2) 2195–2205
Bilous I, Deshko V, Sukhodub I (2018) Parametric analysis of external and internal factors influence
on building energy performance using non-linear multivariate regression models. J Build Eng
20:327–336
Bindra L, Eng K, Ardakanian O, Stroulia E (2022) Flexible, decentralised access control for smart buildings
with smart contracts. Cyber-Phys Syst 8(4): 286–320

13
AI‑big data analytics for building automation and management… 5007

Bode G, Baranski M, Schraven M, Kümpel A, Storek T, Nürenberg M, Müller D, Rothe A, Ziegeldorf JH,
Fütterer J et al (2019) Cloud, wireless technology, internet of things: the next generation of building
automation systems? J Phys 1343:012059
Bode G, Schreiber T, Baranski M, Müller D (2019) A time series clustering approach for building automa-
tion and control systems. Appl Energy 238:1337–1345
Bode G, Thul S, Baranski M, Müller D (2020) Real-world application of machine-learning-based fault
detection trained with experimental data. Energy 198:117323
Bui V, Le NT, Nguyen VH, Kim J, Jang YM et al (2021) Multi-behavior with bottleneck features LSTM for
load forecasting in building energy management system. Electronics 10(9):1026
Bünning F, Huber B, Heer P, Aboudonia A, Lygeros J (2020) Experimental demonstration of data predictive
control for energy optimization and thermal comfort in buildings. Energy Build 211:109792
Cao S-J, Ding J, Ren C (2020) Sensor deployment strategy using cluster analysis of fuzzy c-means algo-
rithm: towards online control of indoor environment’s safety and health. Sustain Cities Soc 59:102190
Carreira P, Costa AA, Mansur V, Arsénio A (2018) Can hvac really learn from users? A simulation-based
study on the effectiveness of voting for comfort and energy use optimization. Sustain Cities Soc
41:275–285
Chemingui Y, Gastli A, Ellabban O (2020) Reinforcement learning-based school energy management sys-
tem. Energies 13(23):6354
Chen Y, Norford LK, Samuelson HW, Malkawi A (2018) Optimal control of HVAC and window systems
for natural ventilation through reinforcement learning. Energy Build 169:195–205
Chen S, Mihara K, Wen J (2018) Time series prediction of CO2, tvoc and hcho based on machine learning
at different sampling points. Build Environ 146:238–246
Chen Y, Tong Z, Zheng Y, Samuelson H, Norford L (2020) Transfer learning with deep neural networks for
model predictive control of hvac and natural ventilation in smart buildings. J Clean Prod 254:119866
Chen K, Chen H, Zhou C, Huang Y, Qi X, Shen R, Liu F, Zuo M, Zou X, Wang J et al (2020) Comparative
analysis of surface water quality prediction performance and identification of key water parameters
using different machine learning models based on big data. Water Res 171:115454
Chen Y, Zhang F, Berardi U (2020) Day-ahead prediction of hourly subentry energy consumption in the
building sector using pattern recognition algorithms. Energy 211:118530
Chen Y, Wen J (2017) Whole building system fault detection based on weather pattern matching and PCA
method. In: 2017 3rd IEEE international conference on control science and systems engineering
(ICCSSE). IEEE, pp 728–732
Choi S, Hur J (2020) An ensemble learner-based bagging model using past output data for photovoltaic
forecasting. Energies 13(6):1438
Choi Y, Yoon S (2021) Autoencoder-driven fault detection and diagnosis in building automation systems:
residual-based and latent space-based approaches. Build Environ 203, 108066
Chou J-S, Ngo N-T (2016) Time series analytics using sliding window metaheuristic optimization-based
machine learning system for identifying building energy consumption patterns. Appl Energy
177:751–770
Chou J-S, Tran D-S (2018) Forecasting energy consumption time series using machine learning techniques
based on usage patterns of residential householders. Energy 165:709–726
Collins AG, Cockburn J (2020) Beyond dichotomies in reinforcement learning. Nat Rev Neurosci
21(10):576–586
Computational Urban Sustainability Platform (CUSP) (n.d.) CUSP—a smart city solution implemented for
Cardiff and Luxembourg. https://​www.​cuspp​latfo​rm.​com/, Accessed 20 Sept 2021
Cortiços ND (2019) Self-learning and self-repairing technologies to establish autonomous building mainte-
nance. In: MATEC Web of conferences, vol 278. EDP Sci, p 04004
Cotrufo N, Zmeureanu R (2016) PCA-based method of soft fault detection and identification for the ongoing
commissioning of chillers. Energy Build 130:443–452
Cruz JCD, Amado TM, Hermogino JQ, Andog MLC, Corpuz FT, Ng JRT, Gonazales JCMB , Inacay JND,
Redoblado JPD, Manuel MCE (2020) Machine learning-based indoor air quality baseline study of the
offices and laboratories of the northwest and southwest building of mapúa university-manila. In: 2020
11th IEEE control and system graduate research colloquium (ICSGRC). IEEE, pp 155–160
Cruz C, Palomar E, Bravo I, Aleixandre M (2021) Behavioural patterns in aggregated demand response
developments for communities targeting renewables. Sustain Cities Soc 72:103001
Culaba AB, Del Rosario AJR, Ubando AT, Chang J-S (2020) Machine learning-based energy consumption
clustering and forecasting for mixed-use buildings. Int J Energy Res 44(12):9659–9673
Das A, Alonso MJ, Mathisen HM et al (2020) Use of deep learning models to predict indoor air quality in a
school case study. In: 16th Conference of the International Society of Indoor Air Quality and Climate:

13
5008 Y. Himeur et al.

creative and smart solutions for better built environments, indoor air 2020. International Society of
Indoor Air Quality and Climate, pp 894–900
de Assis MV, Carvalho LF, Rodrigues JJ, Lloret J, Proença ML Jr (2020) Near real-time security system
applied to SDN environments in IoT networks using convolutional neural network. Comput Electr
Eng 86:106738
de Oliveira EM, Oliveira FLC (2018) Forecasting mid-long term electric energy consumption through
bagging Arima and exponential smoothing methods. Energy 144:776–788
Dehalwar V, Kalam A, Kolhe ML, Zayegh A (2016) Electricity load forecasting for urban area using
weather forecast information. In: 2016 IEEE international conference on power and renewable
energy (ICPRE). IEEE, pp 355–359
Delgado JMD, Oyedele L (2021) Digital twins for the built environment: learning from conceptual and
process models in manufacturing. Adv Eng Inform 49:101332
Delsing J (2017) Local cloud internet of things automation: technology and business model features of
distributed internet of things automation solutions. IEEE Ind Electron Mag 11(4):8–21
Dey M, Rana SP, Dudley S (2020) Smart building creation in large scale HVAC environments through
automated fault detection and diagnosis. Futur Gener Comput Syst 108:950–966
Dey M, Rana SP, Dudley S (2020) A case study based approach for remote fault detection using multi-
level machine learning in a smart building. Smart Cities 3(2):401–419
Diamantoulakis PD, Kapinas VM, Karagiannidis GK (2015) Big data analytics for dynamic energy man-
agement in smart grids. Big Data Res 2(3):94–101
Ding J, Yu CW, Cao S-J (2020) HVAC systems for environmental control to minimize the covid-19
infection. Indoor Built Environ 29(9):1195–1201
Ding X, Du W, Cerpa A (2019) Octopus: deep reinforcement learning for holistic smart building control.
In: Proceedings of the 6th ACM international conference on systems for energy-efficient build-
ings, cities, and transportation, pp 326–335
Do H, Cetin KS (2018) Residential building energy consumption: a review of energy data availabil-
ity, characteristics, and energy performance prediction methods. Curr Sustain Renew Energy Rep
5(1):76–85
Dogruparmak SC, Keskin GA, Yaman S, Alkan A (2014) Using principal component analysis and fuzzy
c-means clustering for the assessment of air quality monitoring. Atmos Pollut Res 5(4):656–663
Doorn N (2021) Artificial intelligence in the water domain: opportunities for responsible use. Sci Total
Environ 755:142561
Dowling CP, Zhang B (2020) Transfer learning for HVAC system fault detection. In: American control
conference (ACC). IEEE 2020:3879–3885
Du B, Zhou Q, Guo J, Guo S, Wang L (2021) Deep learning with long short-term memory neural
networks combining wavelet transform and principal component analysis for daily urban water
demand forecasting. Expert Syst Appl 171:114571
Dun M, Wu L (2020) Forecasting the building energy consumption in china using grey model. Environ
Process 7(3):1009–1022
El Motaki S, Yahyaouy A, Gualous H, Sabor J (2021) A new weighted fuzzy c-means clustering for
workload monitoring in cloud datacenter platforms. Cluster Comput 24, 3367–3379
Elkhoukhi H, NaitMalek Y, Bakhouya M, Berouine A, Kharbouch A, Lachhab F, Hanifi M, El Ouadghiri
D, Essaaidi M (2020) A platform architecture for occupancy detection using stream processing
and machine learning approaches. Concurr Comput Pract Exp 32(17):e5651
Ellis F (1963) The control of operating-suite temperatures. Occup Environ Med 20(4):284–287
Elnour M, Meskin N, Al-Naemi M (2020) Sensor data validation and fault diagnosis using auto-associa-
tive neural network for HVAC systems. J Build Eng 27:100935
Elnour M, Meskin N, Khan K, Jain R (2021) Application of data-driven attack detection framework for
secure operation in smart buildings. Sustain Cities Soc 69:102816
Elnour M, Himeur Y, Fadli F, Mohammedsherif H, Meskin N, Ahmad AM, Petri I, Rezgui Y, Hodorog
A (2022) Neural network-based model predictive control system for optimizing building automa-
tion and management systems of sports facilities. Appl Energy 318:119153
Elnour M, Fadli F, Himeur Y, Petri I, Rezgui Y, Meskin N, Ahmad AM (2022) Performance and energy
optimization of building automation and management systems: towards smart sustainable carbon-
neutral sports facilities. Renew Sustain Energy Rev 162:112401
Elnour M, Meskin N (2022) Novel actuator fault diagnosis framework for multizone hvac systems using
2-D convolutional neural networks. IEEE Trans Automat Sci Eng 19(3) 1985-1996
Elsaeidy A, Munasinghe KS, Sharma D, Jamalipour A (2019) Intrusion detection in smart cities using
restricted Boltzmann machines. J Netw Comput Appl 135:76–83

13
AI‑big data analytics for building automation and management… 5009

El-Sharkawy MF, Javed W (2018) Study of indoor air quality level in various restaurants in Saudi Ara-
bia. Environ Progress Sustain Energy 37(5):1713–1721
energy.gov (2015) An assessment of energy technologies and research opportunities, https://​www.​
energy.​gov/​sites/​prod/​files/​2017/​03/​f34/​qtr-​2015-​chapt​er5.​pdf; Accessed 1 Mar 2021
Englund SM (2007) Safety considerations in the chemical process industries. In: Kent and Riegel’s
handbook of industrial chemistry and biotechnology. Springer, pp 83–146
Estiri H (2014) Building and household x-factors and energy consumption at the residential sector: a struc-
tural equation analysis of the effects of household and building characteristics on the annual energy
consumption of us residential buildings. Energy Econ 43:178–184
Fadli F, Rezgui Y, Petri I, Meskin N, Ahmad AM, Hodorog A, Elnour M, Mohammedsherif H (2021)
Building energy management systems for sports facilities in the gulf region: a focus on impacts and
considerations. CIB
Fan C, Xiao F, Li Z, Wang J (2018) Unsupervised data analytics in mining big building operational data for
energy efficiency enhancement: a review. Energy Build 159:296–308
Fan C, Liu Y, Liu X, Sun Y, Wang J (2021) A study on semi-supervised learning in enhancing performance
of ahu unseen fault detection with limited labeled data. Sustain Cities Soc 70:102874
Fan C, Liu X, Xue P, Wang J (2021) Statistical characterization of semi-supervised neural networks for fault
detection and diagnosis of air handling units. Energy Build 234:110733
Fatema N, Malik H (2021) Data-driven occupancy detection hybrid model using particle swarm optimiza-
tion based artificial neural network. In: Metaheuristic and evolutionary computation: algorithms and
applications. Springer, pp 283–297
Fatema N, Malik H, Iqbal A (2020) Big-data analytics based energy analysis and monitoring for multi-
storey hospital buildings: case study. In: Soft computing in condition monitoring and diagnostics of
electrical and mechanical systems. Springer, pp 325–343
Fathi S, Srinivasan R, Fenner A, Fathi S (2020) Machine learning applications in urban building energy per-
formance forecasting: a systematic review. Renew Sustain Energy Rev 133:110287
Fayed N, Abu-Elkheir M, El-Daydamony E, Atwan A (2019) Sensor-based occupancy detection using neu-
trosophic features fusion. Heliyon 5(9):e02450
Feng C, Mehmani A, Zhang J (2020) Deep learning-based real-time building occupancy detection using ami
data. IEEE Trans Smart Grid 11(5):4490–4501
Feng Y, Duan Q, Chen X, Yakkali SS, Wang J (2021) Space cooling energy usage prediction based on util-
ity data for residential buildings using machine learning methods. Appl Energy 291:116814
Ferdoush Z, Mahmud BN, Chakrabarty A, Uddin J A short-term hybrid forecasting model for time series
electrical-load data using random forest and bidirectional long short-term memory. Int J Electric
Comput Eng (2088-8708) 11(1)
Fernández JMA, Pablo CL (2021) Body temperature and heating temperature in major burns patients care.
Enfermería Global 20(1):478–488
Ferrández-Pastor F-J, Mora H, Jimeno-Morenilla A, Volckaert B (2018) Deployment of IoT edge and fog
computing technologies to develop smart building services. Sustainability 10(11):3832
Francisco A, Mohammadi N, Taylor JE (2020) Smart city digital twin-enabled energy management: toward
real-time urban building energy benchmarking. J Manag Eng 36(2):04019045
Fu F (2020) Fire induced progressive collapse potential assessment of steel framed buildings using machine
learning. J Constr Steel Res 166:105918
Gaboalapswe M (2019) Explore and design an artificial intelligent and data analytic software model to
address domestic water usage billing crisis in botswana urban areas. Ph.D. thesis, Botho University
Gao G, Li J, Wen Y (2020) Deepcomfort: energy-efficient thermal comfort control in buildings via rein-
forcement learning. IEEE Internet Things J 7(9):8472–8484
Gao Y, Ruan Y, Fang C, Yin S (2020) Deep learning and transfer learning models of energy consumption
forecasting for a building with poor information data. Energy Build 223:110156
Gao Y, Fang C, Ruan Y (2019) A novel model for the prediction of long-term building energy demand:
Lstm with attention layer. In: IOP conference series: earth and environmental science, vol 294. IOP
Publishing, p 012033
Giovanis E (2019) Worthy to lose some money for better air quality: applications of Bayesian networks
on the causal effect of income and air pollution on life satisfaction in Switzerland. Empiric Econ
57(5):1579–1611
Gładyszewska-Fiedoruk K, Sulewska MJ (2020) Thermal comfort evaluation using linear discriminant anal-
ysis (lda) and artificial neural networks (anns). Energies 13(3):538
Golabi MR, Radmanesh F, Akhoond-Ali AM, Niksokhan MH, Kisi O (2020) Development of an indirect
method for modelling the water footprint of electricity using wavelet transform coupled with the ran-
dom forest model. Hydrol Sci J 65(15):2521–2534

13
5010 Y. Himeur et al.

Gong M, Bai Y, Qin J, Wang J, Yang P, Wang S (2020) Gradient boosting machine for predicting return
temperature of district heating system: a case study for residential buildings in tianjin. J Build Eng
27:100950
Grillone B, Mor G, Danov S, Cipriano J, Carbonell J, Gabaldón E (2019) Use of generalised additive mod-
els to assess energy efficiency savings in buildings using smart metering data. PROCEEDINGS book
27
Grolinger K, L’Heureux A, Capretz MA, Seewald L (2016) Energy forecasting for event venues: big data
and prediction accuracy. Energy Build 112:222–233
Guo Y, Tan Z, Chen H, Li G, Wang J, Huang R, Liu J, Ahmad T (2018) Deep learning-based fault diag-
nosis of variable refrigerant flow air-conditioning system for building energy saving. Appl Energy
225:732–745
Hafeez G, Alimgeer KS, Khan I (2020) Electric load forecasting based on deep learning and optimized by
heuristic algorithm in smart grid. Appl Energy 269:114915
Haidar N, Tamani N, Nienaber F, Wesseling MT, Bouju A, Ghamri-Doudane Y, (2019) Data collection
period and sensor selection method for smart building occupancy prediction. In: IEEE 89th vehicular
technology conference (VTC2019-Spring). IEEE, pp. 1–6
Hall S, Cooper WE, Marciani L, McGee JM (2011) Security management for sports and special events: an
interagency approach to creating safe facilities. Hum Kinet
Han H, Cui X, Fan Y, Qing H (2019) Least squares support vector machine (LS-SVM)-based chiller fault
diagnosis using fault indicative features. Appl Therm Eng 154:540–547
Han H, Zhang Z, Cui X, Meng Q (2020) Ensemble learning with member optimization for fault diagnosis of
a building energy system. Energy Build 226:110351
Haq IU, Ullah A, Khan SU, Khan N, Lee MY, Rho S, Baik SW (2021) Sequential learning-based energy
consumption prediction model for residential and commercial sectors. Mathematics 9(6):605
Hasanzadeh Nafari R, Ngo T, Mendis P (2016) An assessment of the effectiveness of tree-based models for
multi-variate flood damage assessment in Australia. Water 8(7):282
He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition. In: Proceedings of the
IEEE conference on computer vision and pattern recognition, pp 770–778
Himeur Y, Alsalemi A, Bensaali F, Amira A (2020) A novel approach for detecting anomalous energy con-
sumption based on micro-moments and deep neural networks. Cogn Comput 12(6):1381–1401
Himeur Y, Alsalemi A, Bensaali F, Amira A (2020) Effective non-intrusive load monitoring of buildings
based on a novel multi-descriptor fusion with dimensionality reduction. Appl Energy 279:115872
Himeur Y, Alsalemi A, Al-Kababji A, Bensaali F, Amira A (2020) Data fusion strategies for energy effi-
ciency in buildings: overview, challenges and novel orientations. Inform Fus 64:99–120
Himeur Y, Alsalemi A, Bensaali F, Amira A (2020) Building power consumption datasets: survey, taxon-
omy and future directions. Energy Build 227:110404
Himeur Y, Alsalemi A, Al-Kababji A, Bensaali F, Amira A, Sardianos C, Dimitrakopoulos G, Varlamis I
(2021) A survey of recommender systems for energy efficiency in buildings: principles, challenges
and prospects. Inform Fus 72:1–21
Himeur Y, Ghanem K, Alsalemi A, Bensaali F, Amira A (2021) Artificial intelligence based anomaly detec-
tion of energy consumption in buildings: a review, current trends and new perspectives. Appl Energy
287:116601
Himeur Y, Alsalemi A, Bensaali F, Amira A (2021) Smart power consumption abnormality detection in
buildings using micromoments and improved k-nearest neighbors. Int J Intell Syst 36(6):2865–2894
Himeur Y, Alsalemi A, Bensaali F, Amira A (2021) An intelligent nonintrusive load monitoring scheme
based on 2d phase encoding of power signals. Int J Intell Syst 36(1):72–93
Himeur Y, Alsalemi A, Bensaali F, Amira A (2021) Smart non-intrusive appliance identification using a
novel local power histogramming descriptor with an improved k-nearest neighbors classifier. Sustain
Cities Soc 67:102764
Himeur Y, Alsalemi A, Bensaali F, Amira A, Varlamis I, Bravos G, Sardianos C, Dimitrakopoulos G (2022)
Techno-economic assessment of building energy efficiency systems using behavioral change: a case
study of an edge-based micro-moments solution. J Clean Prod 331:129786
Himeur Y, Sayed A, Alsalemi A, Bensaali F, Amira A, Varlamis I, Eirinaki M, Sardianos C, Dimitrakopou-
los G (2022) Blockchain-based recommender systems: applications, challenges and future opportuni-
ties. Comput Sci Rev 43:100439
Himeur Y, Alsalemi A, Bensaali F, Amira A (2021) Appliance identification using a histogram post-pro-
cessing of 2d local binary patterns for smart grid applications. In: 2020 25th International conference
on pattern recognition (ICPR). IEEE, pp 5744–5751

13
AI‑big data analytics for building automation and management… 5011

Himeur Y, Alsalemi A, Bensaali F, Amira A (2021) The emergence of hybrid edge-cloud computing for
energy efficiency in buildings. In: Proceedings of SAI intelligent systems conference. Springer, pp
70–83
Himeur Y, Alsalemi A, Bensaali F, Amira A (2022) Detection of appliance-level abnormal energy con-
sumption in buildings using autoencoders and micro-moments. In: International conference on big
data and internet of things. Springer, pp 179–193
Himeur Y, Alsalemi A, Bensaali F, Amira A, Al-Kababji (2022) A Recent trends of smart nonintrusive load
monitoring in buildings: a review, open challenges, and future directions. Int J Intell Syst 37(10):
7124–7179
Himeur Y, Alsalemi A, Bensaali F, Amira A, Varlamis I, Bravos G, Sardianos C, Dimitrakopoulos G Mar-
ketability of building energy efficiency systems based on behavioral change: a case study of a novel
micro-moments based solution. arXiv:​2105.​10460
Himeur Y, Elnour M, Fadli F, Meskin N, Petri I, Rezgui Y, Bensaali F, Amira A (2022) Next-generation
energy systems for sustainable smart cities: roles of transfer learning. Sustain Cities Soc 104059
Himeur Y, Sohail SS, Bensaali F, Amira A, Alazab M (2022) Latest trends of security and privacy in recom-
mender systems: a comprehensive review and future perspectives. Comput Security 102746
Hu J, Vasilakos AV (2016) Energy big data analytics and security: challenges and opportunities. IEEE Trans
Smart Grid 7(5):2423–2436
Hu H, Wang L, Peng L, Zeng Y-R (2020) Effective energy consumption forecasting using enhanced bagged
echo state network. Energy 193:116778
Huang Q, Hao K (2020) Development of CNN-based visual recognition air conditioner for smart buildings.
J Inf Technol Constr 25:361–373
Huang S, Zuo W, Sohn MD (2018) A Bayesian network model for predicting cooling load of commercial
buildings. In: Building simulation, vol 11. Springer, pp 87–101
Huchuk B, Sanner S, O’Brien W (2019) Comparison of machine learning models for occupancy prediction
in residential buildings using connected thermostat data. Build Environ 160:106177
Huda AN, Taib S, Jadin MS, Ishak D (2012) A semi-automatic approach for thermographic inspection of
electrical installations within buildings. Energy Build 55:585–591
Idowu S, Saguna S, Åhlund C, Schelén O (2016) Applied machine learning: forecasting heat load in district
heating system. Energy Build 133:478–488
IoT Security Foundation, Smart cities—the emergence of the CyberSafe building. IoT Security Foundation.
https://​www.​iotse​curit​yfoun​dation.​org/​smart_​cities_​the_​emerg​ence_​of_​the_​cyb er_​safe_​build​ing/.
Accessed 28 June 2020
Ippolito M, Riva Sanseverino E, Zizzo G (2014) Impact of building automation control systems and techni-
cal building management systems on the energy performance class of residential buildings: an Italian
case study. Energy Build 69:33–40
Ishaq M, Kwon S et al (2021) Short-term energy forecasting framework using an ensemble deep learning
approach. IEEE Access 9:94262–94271
Janarthanan R, Partheeban P, Somasundaram K, Elamparithi PN (2021) A deep learning approach for pre-
diction of air quality index in a metropolitan city. Sustain Cities Soc 67:102720
Javadzadeh G, Rahmani AM (2020) Fog computing applications in smart cities: a systematic survey. Wire-
less Netw 26(2):1433–1457
Jenny H, Wang Y, Alonso EG, Minguez R (2020) Using artificial intelligence for smart water management
systems. Asian Development Bank. http://​hdl.​handle.​net/​11540/​12225.
Ji T, Liu L, Wang T, Lin W, Li M, Wu Q (2019) Non-intrusive load monitoring using additive facto-
rial approximate maximum a posteriori based on iterative fuzzy c-means. IEEE Trans Smart Grid
10(6):6667–6677
Jia M, Komeily A, Wang Y, Srinivasan RS (2019) Adopting internet of things for the development of smart
buildings: a review of enabling technologies and applications. Autom Constr 101:111–126
Juntarawijit C, Juntarawijit Y (2017) Cooking smoke and respiratory symptoms of restaurant workers in
Thailand. BMC Pulm Med 17(1):1–11
Kaklauskas A, Lill I, Amaratunga D, Ubarte I (2019) Model for smart, self-learning and adaptive resilience
building. In: 10th Nordic conference on construction economics and organization, Emerald Publish-
ing Limited
Kalajdjieski J, Zdravevski E, Corizzo R, Lameski P, Kalajdziski S, Pires IM, Garcia NM, Trajkovik
V (2020) Air pollution prediction with multi-modal data and deep neural networks. Remote Sens
12(24):4142
Kalantzis V, Kollias G, Ubaru S, Nikolakopoulos AN, Horesh L, Clarkson K (2021) Projection techniques
to update the truncated svd of evolving matrices with applications. In: International conference on
machine learning, PMLR, pp 5236–5246

13
5012 Y. Himeur et al.

Kallio J, Tervonen J, Räsänen P, Mäkynen R, Koivusaari J, Peltola J (2021) Forecasting office indoor CO2
concentration using machine learning with a one-year dataset. Build Environ 187:107409
Kang T, Chen P, Quackenbush J, Ding W (2020) A novel deep learning model by stacking conditional
restricted boltzmann machine and deep neural network. In: Proceedings of the 26th ACM SIGKDD
international conference on knowledge discovery & data mining, pp 1316–1324
Kaspersky, Nearly four in ten smart buildings targeted by malicious attacks in H1 2019, https://www.usa.
kaspersky.com/about/press-releases/2019_smart-buildings-threat-landscape/, Accessed on 28 June
2020 (2019)
Katipamula S (2019) Building automation: where is it today and where it should be. In: CASE, p 1
Khalil M, McGough S, Pourmirza Z, Pazhoohesh M, Walker S, Transfer learning approach for occupancy
prediction in smart buildings. In: (2021) 12th International renewable engineering conference (IREC).
IEEE 2021, pp 1–6
Khamma TR, Zhang Y, Guerrier S, Boubekri M (2020) Generalized additive models: an efficient method
for short-term energy prediction in office buildings. Energy 213:118834
Khan LU, Yaqoob I, Tran NH, Kazmi SA, Dang TN, Hong CS (2020) Edge-computing-enabled smart cities:
a comprehensive survey. IEEE Internet Things J 7(10):10200–10232
Khan ZA, Ullah A, Ullah W, Rho S, Lee M, Baik SW (2020) Electrical energy prediction in residential
buildings for short-term horizons using hybrid deep learning strategy. Appl Sci 10(23):8634
Khan AN, Iqbal N, Ahmad R, Kim D-H (2021) Ensemble prediction approach based on learning to statisti-
cal model for efficient building energy consumption management. Symmetry 13(3):405
Khattak HA, Farman H, Jan B, Din IU (2019) Toward integrating vehicular clouds with IoT for smart city
services. IEEE Network 33(2):65–71
Khwaja A, Naeem M, Anpalagan A, Venetsanopoulos A, Venkatesh B (2015) Improved short-term load
forecasting using bagged neural networks. Electric Power Syst Res 125:109–115
Kiliccote S, Piette MA, Hansen D Advanced controls and communications for demand response and energy
efficiency in commercial buildings
Kim R, Hong Y, Choi Y, Yoon S (2021) System-level fouling detection of district heating substations using
virtual-sensor-assisted building automation system. Energy 227:120515
Koo J, Yoon S, Kim J (2022) Virtual in situ calibration for operational backup virtual sensors in building
energy systems. Energies 15(4):1394
Krause B, Lu L, Murray I, Renals S Multiplicative LSTM for sequence modelling. arXiv:​1609.​07959
Krizhevsky A, Sutskever I, Hinton GE (2012) ImageNet classification with deep convolutional neural net-
works. Adv Neural Inf Process Syst 25:1097–1105
Kučera A, Pitner T (2018) Semantic bms: allowing usage of building automation data in facility benchmark-
ing. Adv Eng Inform 35:69–84
Lagesse B, Wang S, Larson TV, Kim AA (2020) Predicting pm2.5 in well-mixed indoor air for a large
office building using regression and artificial neural network models. Environ Sci Technol
54(23):15320–15328
Lee S, Karava P (2020) Towards smart buildings with self-tuned indoor thermal environments—a critical
review. Energy Build 224:110172
Li M (2020) Optimizing HVAC, systems in buildings with machine learning prediction models: an algo-
rithm based economic analysis. In: 2020 Management science informatization and economic innova-
tion development conference (MSIEID). IEEE, pp 210–217
Li Y, Tong Z (2021) Model predictive control strategy using encoder-decoder recurrent neural networks for
smart control of thermal environment. J Build Eng 42:103017
Li S, Wen J (2014) A model-based fault detection and diagnostic methodology based on PCA method and
wavelet transform. Energy Build 68:63–71
Li W, Logenthiran T, Phan V-T, Woo WL (2018) Implemented iot-based self-learning home management
system (shms) for Singapore. IEEE Internet Things J 5(3):2212–2219
Li Z, Friedrich D, Harrison GP (2020) Demand forecasting for a mixed-use building using agent-schedule
information with a data-driven model. Energies 13(4):780
Li B, Cheng F, Zhang X, Cui C, Cai W (2021) A novel semi-supervised data-driven method for chiller fault
diagnosis with unlabeled data. Appl Energy 285:116459
Li B, Cheng F, Cai H, Zhang X, Cai W (2021) A semi-supervised approach to fault detection and diagno-
sis for building HVAC systems based on the modified generative adversarial network. Energy Build
246:111044
Liang P, Yang H-D, Chen W-S, Xiao S-Y, Lan Z-Z (2018) Transfer learning for aluminium extrusion
electricity consumption anomaly detection via deep neural networks. Int J Comput Integr Manuf
31(4–5):396–405

13
AI‑big data analytics for building automation and management… 5013

Lin F, Liu S-J, Chao H-C, Pan J-S (2021) Short-term household load forecasting model based on variational
mode decomposition and gated recurrent unit with attention mechanism. J Netw Intell 6(1):143–153
Li L, Ota K, Dong M (2017) Everything is image: Cnn-based short-term electrical load forecasting for
smart grid. In: 2017 14th International symposium on pervasive systems, algorithms and networks
& 2017 11th International conference on frontier of computer science and technology & 2017 Third
international symposium of creative computing (ISPAN-FCST-ISCC). IEEE, pp 344–351
Lissa P, Deane C, Schukat M, Seri F, Keane M, Barrett E (2021) Deep reinforcement learning for home
energy management system control. Energy AI 3:100043
Liu G, Yang J, Hao Y, Zhang Y (2018) Big data-informed energy efficiency assessment of china industry
sectors based on k-means clustering. J Clean Prod 183:304–314
Liu N, Liu X, Jayaratne R, Morawska L (2020) A study on extending the use of air quality monitor data
via deep learning techniques. J Clean Prod 274:122956
Liu J, Wang Y, Zhang Y (2020) A novel isomap-SVR soft sensor model and its application in rotary kiln
calcination zone temperature prediction. Symmetry 12(1):167
Liu J, Zhang Q, Li X, Li G, Liu Z, Xie Y, Li K, Liu B (2021) Transfer learning-based strategies for fault
diagnosis in building energy systems. Energy Build 250:111256
Liu Z, Chi Z, Osmani M, Demian P (2021) Blockchain and building information management (bim) for
sustainable building development within the context of smart cities. Sustainability 13(4):2090
Li A, Xiao F, Fan C, Hu M (2021) Development of an ann-based building energy model for information-
poor buildings using transfer learning. In: Building simulation, vol 14. Springer, pp 89–101
Lobdell KW, Stamou S, Sanchez JA (2012) Hospital-acquired infections. Surg Clin 92(1):65–77
Lopes MADS, Neto ADD, Martins ADM (2020) Parallel t-sne applied to data visualization in smart cit-
ies. IEEE Access 8:11482–11490
Loy-Benitez J, Heo S, Yoo C (2020) Imputing missing indoor air quality data via variational convo-
lutional autoencoders: implications for ventilation management of subway metro systems. Build
Environ 182:107135
Loy-Benitez J, Heo S, Yoo C (2020) Soft sensor validation for monitoring and resilient control of
sequential subway indoor air quality through memory-gated recurrent neural networks-based
autoencoders. Control Eng Pract 97:104330
Loy-Benitez J, Li Q, Nam K, Yoo C (2020) Sustainable subway indoor air quality monitoring and fault-
tolerant ventilation control using a sparse autoencoder-driven sensor self-validation. Sustain Cities
Soc 52:101847
Lu TT (2009) Fundamental limitations of semi-supervised learning. Master’s thesis, University of
Waterloo
Lu H, Cheng F, Ma X, Hu G (2020) Short-term prediction of building energy consumption employing an
improved extreme gradient boosting model: a case study of an intake tower. Energy 203:117756
Lv Z, Qiao L, Kumar Singh A, Wang Q (2021) Ai-empowered IoT security for smart cities. ACM Trans
Internet Technol 21(4):1–21
Ma J, Cheng JC, Lin C, Tan Y, Zhang J (2019) Improving air quality prediction accuracy at larger tem-
poral resolutions using deep learning and transfer learning techniques. Atmos Environ 214:116885
Maatoug A, Belalem G, Mahmoudi S (2019) Fog computing framework for location-based energy man-
agement in smart buildings. Multiagent Grid Syst 15(1):39–56
Mad Saad S, Andrew AM, Shakaff AY, Mat Dzahir MA, Hussein M, Mohamad M, Ahmad ZA (2017)
Pollutant recognition based on supervised machine learning for indoor air quality monitoring sys-
tems. Appl Sci 7(8):823
Mahmud MS, Huang JZ, Salloum S, Emara TZ, Sadatdiynov K (2020) A survey of data partitioning and
sampling methods to support big data analysis. Big Data Min Anal 3(2):85–101. https://​doi.​org/​
10.​26599/​BDMA.​2019.​90200​15
Mansor AA, Shamsul S, Abdullah S, Dom N, Napi NM, Ahmed A, Ismail M (2021) Identification of
indoor air quality (iaq) sources in libraries through principal component analysis (PCA). In: IOP
conference series: materials science and engineering, vol 1144. IOP Publishing, p 012055
Mariano-Hernández D, Hernández-Callejo L, Zorita-Lamadrid A, Duque-Pérez O, García FS (2021) A
review of strategies for building energy management system: model predictive control, demand
side management, optimization, and fault detect & diagnosis. J Build Eng 33:101692
Marino DL, Amarasinghe K, Manic M (2016) Building energy load forecasting using deep neural net-
works. In: IECON 2016-42nd annual conference of the IEEE Industrial Electronics Society. IEEE,
pp 7046–7051
Markoska E, Lazarova-Molnar S (2018) Towards smart buildings performance testing as a service. In: 2018
Third international conference on fog and mobile edge computing (FMEC). IEEE, pp 277–282

13
5014 Y. Himeur et al.

Mawson VJ, Hughes BR (2020) Deep learning techniques for energy forecasting and condition monitor-
ing in the manufacturing sector. Energy Build 217:109966
Merz H, Hansemann T, Hübner C (2018) Building automation: communication systems with EIB/KNX,
Lon and BACnet, 2nd edn. Springer, Cham
Miori V, Russo D, Ferrucci L (2019) Interoperability of home automation systems as a critical chal-
lenge for IoT. In: 2019 4th International conference on computing, communications and security
(ICCCS). IEEE, pp 1–7
Mo H, Sun H, Liu J, Wei S (2019) Developing window behavior models for residential buildings using
xgboost algorithm. Energy Build 205:109564
Mohamed N, Al-Jaroodi J, Lazarova-Molnar S (2018) Energy cloud: services for smart buildings. In:
Sustainable cloud and energy services. Springer, pp 117–134
Moher D, Liberati A, Tetzlaff J, Altman DG, Group P (2009) Preferred reporting items for systematic
reviews and meta-analyses: the prisma statement. PLoS Med 6(7):e1000097
Molinara M, Ferdinandi M, Cerro G, Ferrigno L, Massera E (2020) An end to end indoor air monitoring
system based on machine learning and sensiplus platform. IEEE Access 8:72204–72215
Molina-Solana M, Ros M, Ruiz MD, Gómez-Romero J, Martín-Bautista MJ (2017) Data science for
building energy management: a review. Renew Sustain Energy Rev 70:598–609
Moon J, Park J, Hwang E, Jun S (2018) Forecasting power consumption for higher educational institu-
tions based on machine learning. J Supercomput 74(8):3778–3800
Moon J, Park S, Rho S, Hwang E (2019) A comparative analysis of artificial neural network architectures
for building energy consumption forecasting. Int J Distrib Sens Netw 15(9):1550147719877616
Moradzadeh A, Mansour-Saatloo A, Mohammadi-Ivatloo B, Anvari-Moghaddam A (2020) Performance
evaluation of two machine learning techniques in heating and cooling loads forecasting of residen-
tial buildings. Appl Sci 10(11):3829
Movahedi A, Derrible S Interrelationships between electricity, gas, and water consumption in large-scale
buildings. J Ind Ecol
Mtibaa F, Nguyen K-K, Dermardiros V, Cheriet M (2021) Context-aware model predictive control
framework for multi-zone buildings. J Build Eng 42:102340
Muhammad S, Sapri M, Sipan I (2014) Academic buildings and their influence on students’ wellbeing in
higher education institutions. Soc Indic Res 115(3):1159–1178
Muiruri D et al (2021) Modelling indoor air quality using sensor data and machine learning methods
Mukherjee P, Barik R, Pradhan C (2021) echain: Leveraging toward blockchain technology for smart
energy utilization. In: Applications of advanced computing in systems. Springer, pp 73–81
Mumtaz R, Zaidi SMH, Shakir MZ, Shafi U, Malik MM, Haque A, Mumtaz S, Zaidi SAR (2021) Inter-
net of things (IoT) based indoor air quality sensing and predictive analytic-a covid-19 perspective.
Electronics 10(2):184
Mundt T, Dähn A, Glock H-W (2014) Forensic analysis of home automation systems. In: 7th Workshop
on hot topics in privacy enhancing technologies (HotPETs 2014)
Muntean M, Dănăiaţă D, Hurbean L, Jude C (2021) A business intelligence & analytics framework for
clean and affordable energy data analysis. Sustainability 13(2):638
Mutis I, Ambekar A, Joshi V (2020) Real-time space occupancy sensing and human motion analysis
using deep learning for indoor air quality control. Autom Constr 116:103237
Nasser AA, Rashad MZ, Hussein SE (2020) A two-layer water demand prediction system in urban areas
based on micro-services and LSTM neural networks. IEEE Access 8:147647–147661
Nawari NO, Ravindran S (2019) Blockchain and the built environment: potentials and limitations. J
Build Eng 25:100832
Nejat P, Hussen HM, Fadli F, Chaudhry HN, Calautit J, Jomehzadeh F (2020) Indoor environmental
quality (ieq) analysis of a two-sided windcatcher integrated with anti-short-circuit device for low
wind conditions. Processes 8(7):840
Nguyen VK, Zhang WE, Mahmood A (2021) Semi-supervised intrusive appliance load monitoring
in smart energy monitoring system. ACM Trans Multimed Comput Commun Appl (TOMM)
17(3):1–20
Nie P, Roccotelli M, Fanti MP, Ming Z, Li Z (2021) Prediction of home energy consumption based on
gradient boosting regression tree. Energy Rep 7:1246–1255
O’Grady T, Chong H-Y, Morrison GM (2021) A systematic review and meta-analysis of building auto-
mation systems. Build Environ 195:107770
Ouache R, Nahiduzzaman KM, Hewage K, Sadiq R (2021) Performance investigation of fire protection
and intervention strategies: artificial neural network-based assessment framework. J Build Eng
42:102439

13
AI‑big data analytics for building automation and management… 5015

Ozturk GB (2020) Interoperability in building information modeling for aeco/fm industry. Autom Constr
113:103122
Pal N, Ghosh P, Karsai G (2019) DeepECO: applying deep learning for occupancy detection from energy
consumption data. In: 18th IEEE international conference on machine learning and applications
(ICMLA). IEEE, pp 1938–1943
Park JY, Yang X, Miller C, Arjunan P, Nagy Z (2019) Apples or oranges? Iidentification of fundamental
load shape profiles for benchmarking buildings using a large and diverse dataset. Appl Energy
236:1280–1295
Park S, Moon J, Jung S, Rho S, Baik SW, Hwang E (2020) A two-stage industrial load forecasting
scheme for day-ahead combined cooling, heating and power scheduling. Energies 13(2):443
Park S, Jung S, Jung S, Rho S, Hwang E (2021) Sliding window-based lightgbm model for electric load
forecasting using anomaly repair. J Supercomput 1–22
Pathak N, Ba A, Ploennigs J, Roy N (2018) Forecasting gas usage for big buildings using generalized
additive models and deep learning. In: 2018 IEEE international conference on smart computing
(SMARTCOMP). IEEE, pp 203–210
Pešić S, Tošić M, Iković O, Radovanović M, Ivanović M, Bošković D (2019) BLEMAT: data analytics
and machine learning for smart building occupancy detection and prediction. Int J Artif Intell
Tools 28(06):1960005
Petri I, Rana O, Rezgui Y, Fadli F Edge HVAC analytics. Energies 14(17)
Pham A-D, Ngo N-T, Truong TTH, Huynh N-T, Truong N-S (2020) Predicting energy consumption in
multiple buildings using machine learning for improving energy efficiency and sustainability. J
Clean Prod 260:121082
Pinto T, Praça I, Vale Z, Silva J (2021) Ensemble learning for electricity consumption forecasting in
office buildings. Neurocomputing 423:747–755
Pinto G, Wang Z, Roy A, Hong T, Capozzoli A (2022) Transfer learning for smart buildings: a critical
review of algorithms, applications, and future perspectives. Adv Appl Energy 100084
Plageras AP, Psannis KE, Stergiou C, Wang H, Gupta BB (2018) Efficient IoT-based sensor big data
collection-processing and analysis in smart buildings. Futur Gener Comput Syst 82:349–357
Png E, Srinivasan S, Bekiroglu K, Chaoyang J, Su R, Poolla K (2019) An internet of things upgrade for
smart and scalable heating, ventilation and air-conditioning control in commercial buildings. Appl
Energy 239:408–424
Pratt BW, Erickson JD (2020) Defeat the peak: behavioral insights for electricity demand response pro-
gram design. Energy Res Soc Sci 61:101352
Pulimeno M, Piscitelli P, Colazzo S, Colao A, Miani A (2020) Indoor air quality at school and students’
performance: Recommendations of the unesco chair on health education and sustainable develop-
ment & the Italian society of environmental medicine (sima). Health Promot Perspect 10(3):169
Qin J, Zhang J (2017) Sampling for building energy consumption with fuzzy theory. Energy Build
156:78–84
Quinn C, Shabestari AZ, Misic T, Gilani S, Litoiu M, McArthur J (2020) Building automation system-
bim integration using a linked data structure. Autom Constr 118:103257
Rahim MS, Nguyen KA, Stewart RA, Giurco D, Blumenstein M (2020) Machine learning and data ana-
lytic techniques in digital water metering: a review. Water 12(1):294
Ray PP, Dash D, De D (2019) Edge computing for internet of things: a survey, e-healthcare case study
and future direction. J Netw Comput Appl 140:1–22
Razavi R, Gharipour A, Fleury M, Akpan IJ (2019) Occupancy detection of residential buildings using
smart meter data: a large-scale study. Energy Build 183:195–208
Rehman SU, Javed AR, Khan MU, Nazar Awan M, Farukh A, Hussien A (2020) Personalisedcomfort:
a personalised thermal comfort model to predict thermal sensation votes for smart building resi-
dents. Enterprise Inform Syst 1–23
Ribeiro M, Grolinger K, ElYamany HF, Higashino WA, Capretz MA (2018) Transfer learning with sea-
sonal and trend adjustment for cross-building energy forecasting. Energy Build 165:352–363
Roccetti M, Delnevo G, Casini L, Cappiello G (2019) Is bigger always better? a controversial journey
to the center of machine learning design, with uses and misuses of big data for predicting water
meter failures. J Big Data 6(1):1–23
Rocha Filho GP, Mano LY, Valejo ADB, Villas LA, Ueyama J (2018) A low-cost smart home automa-
tion to enhance decision-making based on fog computing and computational intelligence. IEEE
Lat Am Trans 16(1):186–191
Roelofsen P The impact of office environments on employee performance: The design of the workplace
as a strategy for productivity enhancement. J Facilit Manag

13
5016 Y. Himeur et al.

Roger Rozario A et al (2021) Forecasting-mining prediction of water consumption for residential sec-
tors. Ann Roman Soc Cell Biol 25(6):2918–2924
Runge J, Zmeureanu R A review of deep learning techniques for forecasting energy use in buildings.
Energies 14(3)
Saini J, Dutta M, Marques G (2020) Indoor air quality monitoring systems based on internet of things: a
systematic review. Int J Environ Res Public Health 17(14):4942
Saini J, Dutta M, Marques G (2020) A comprehensive review on indoor air quality monitoring systems
for enhanced public health. Sustain Environ Res 30(1):1–12
Sajjad M, Khan ZA, Ullah A, Hussain T, Ullah W, Lee MY, Baik SW (2020) A novel cnn-gru-based
hybrid approach for short-term residential load forecasting. IEEE Access 8:143759–143768
Salerno VM, Rabbeni G (2018) An extreme learning machine approach to effective energy disaggrega-
tion. Electronics 7(10):235
Salonen H, Lahtinen M, Lappalainen S, Nevala N, Knibbs LD, Morawska L, Reijula K (2013) Physical
characteristics of the indoor environment that affect health and wellbeing in healthcare facilities: a
review. Intell Build Int 5(1):3–25
Sardianos C, Varlamis I, Dimitrakopoulos G, Anagnostopoulos D, Alsalemi A, Bensaali F, Himeur Y,
Amira A (2020) Rehab-c: recommendations for energy habits change. Futur Gener Comput Syst
112:394–407
Sardianos C, Varlamis I, Chronis C, Dimitrakopoulos G, Alsalemi A, Himeur Y, Bensaali F, Amira A
(2021) The emergence of explainability of intelligent systems: delivering explainable and personal-
ized recommendations for energy efficiency. Int J Intell Syst 36(2):656–680
Sardianos C, Chronis C, Varlamis I, Dimitrakopoulos G, Himeur Y, Alsalemi A, Bensaali F, Amira A
(2020) Real-time personalised energy saving recommendations. In International conferences on inter-
net of things (iThings) and IEEE green computing and communications (GreenCom) and IEEE cyber,
physical and social computing (CPSCom) and IEEE smart data (SmartData) and IEEE congress on
cybermatics (Cybermatics). IEEE, pp 366–371
Sardianos C, Varlamis I, Chronis C, Dimitrakopoulos G, Himeur Y, Alsalemi A, Bensaali F, Amira A
(2020) A model for predicting room occupancy based on motion sensor data. In: 2020 IEEE interna-
tional conference on informatics, IoT, and enabling technologies (ICIoT). IEEE, pp 394–399
Sayed AN, Himeur Y, Bensaali F (2022) Deep and transfer learning for building occupancy detection: a
review and comparative analysis. Eng Appl Artif Intell 115:105254
Sayed A, Alsalemi A, Himeur Y, Bensaali F, Amira A (2022) Endorsing energy efficiency through accurate
appliance-level power monitoring, automation and data visualization. In: Networking, intelligent sys-
tems and security. Springer, pp 603–617
Sayed A, Himeur Y, Alsalemi A, Bensaali F, Amira A Intelligent edge-based recommender system for inter-
net of energy applications. IEEE Syst J
Schachinger D, Fernbach A, Kastner W (2017) Modeling framework for IoT integration of building automa-
tion systems. At-Automatisierungstechnik 65(9):630–640
Serale G, Fiorentini M, Capozzoli A, Bernardini D, Bemporad A (2018) Model predictive control (mpc) for
enhancing building and hvac system energy efficiency: problem formulation, applications and oppor-
tunities. Energies 11(3):631
Seyedzadeh S, Rahimian FP, Glesk I, Roper M (2018) Machine learning for estimation of building energy
consumption and performance: a review. Visual Eng 6(1):1–20
Seyedzadeh S, Pour Rahimian F, Rastogi P, Glesk I (2019) Tuning machine learning models for prediction
of building energy loads. Sustain Cities Soc 47:101484
Sha H, Xu P, Hu C, Li Z, Chen Y, Chen Z (2019) A simplified HVAC energy prediction method based on
degree-day. Sustain Cities Soc 51:101698
Shahnazari H, Mhaskar P, House JM, Salsbury TI (2019) Modeling and fault diagnosis design for HVAC
systems using recurrent neural networks. Comput Chem Eng 126:189–203
Shahzad M, Shafiq MT, Douglas D, Kassem M (2022) Digital twins in built environments: an investigation
of the characteristics, applications, and challenges. Buildings 12(2):120
Sharma PK, Mondal A, Jaiswal S, Saha M, Nandi S, De T, Saha S (2021) Indoairsense: a framework for
indoor air quality estimation and forecasting. Atmos Pollut Res 12(1):10–22
Sharma A, Sabitha AS, Bansal A (2018) Edge analytics for building automation systems: a review. In: 2018
International conference on advances in computing, communication control and networking (ICAC-
CCN). IEEE, pp 585–590
Shine P, Murphy MD, Upton J, Scully T (2018) Machine-learning algorithms for predicting on-farm direct
water and electricity consumption on pasture based dairy farms. Comput Electron Agric 150:74–87
Sideratos G, Ikonomopoulos A, Hatziargyriou ND (2020) A novel fuzzy-based ensemble model for load
forecasting using hybrid deep neural networks. Electric Power Syst Res 178:106025

13
AI‑big data analytics for building automation and management… 5017

Singh S, Yassine A (2018) Big data mining of energy time series for behavioral analytics and energy con-
sumption forecasting. Energies 11(2):452
Siountri K, Skondras E, Vergados DD (2020) Developing smart buildings using blockchain, internet of
things, and building information modeling. Int J Interdiscipli Telecommun Netw (IJITN) 12(3):1–15
Skomski E, Lee J-Y, Kim W, Chandan V, Katipamula S, Hutchinson B (2020) Sequence-to-sequence neu-
ral networks for short-term electrical load forecasting in commercial office buildings. Energy Build
226:110350
Smolak K, Kasieczka B, Fialkiewicz W, Rohm W, Siła-Nowicka K, Kopańczyk K (2020) Applying human
mobility and water consumption data for short-term water demand forecasting using classical and
machine learning models. Urban Water J 17(1):32–42
Somontina JAB, Garcia FCC, Macabebe EQB (2018) Water consumption monitoring with fixture recogni-
tion using random forest. In: TENCON 2018-2018 IEEE region 10 conference. IEEE, pp 0663–0667
Somu N, Gauthama Raman MR, Ramamritham K (2020) A hybrid model for building energy consumption
forecasting using long short term memory networks. Appl Energy 261:114131
Somu N, Gauthama Raman MR, Ramamritham K (2021) A deep learning framework for building energy
consumption forecast. Renew Sustain Energy Rev 137:110591
Stamatescu G, Stamatescu I, Arghira N, Făgărăşan I (2020) Cybersecurity perspectives for smart building
automation systems. In: 2020 12th International conference on electronics, computers and artificial
intelligence (ECAI). IEEE, pp 1–5
Stergiou C, Psannis KE, Gupta BB, Ishibashi Y (2018) Security, privacy & efficiency of sustainable cloud
computing for big data & IoT. Sustain Comput Inform Syst 19:174–184
Su B, Wang S (2020) An agent-based distributed real-time optimal control strategy for building HVAC
systems for applications in the context of future IoT-based smart sensor networks. Appl Energy
274:115322
Sultan MM, Islam MS, Rahman MA (2017) Smart fire detection system with early notifications using
machine learning. Int J Comput Intell Appl 16(02):1750009
Sun AY, Scanlon BR (2019) How can big data and machine learning benefit environment and water man-
agement: a survey of methods, applications, and future directions. Environ Res Lett 14(7):073001
Sun F, Yu J (2021) Improved energy performance evaluating and ranking approach for office buildings using
simple-normalization, entropy-based topsis and k-means method. Energy Rep 7:1560–1570
Sun C, Zhai Z (2020) The efficacy of social distance and ventilation effectiveness in preventing covid-19
transmission. Sustain Cities Soc 62:102390
Sun Y, Haghighat F, Fung BC (2020) A review of the-state-of-the-art in data-driven approaches for building
energy prediction. Energy Build 221:110022
Sun L, Wei Q, He L, Yin Z (2020) The prediction of building heating and ventilation energy consumption
base on adaboost-bp algorithm. In: IOP Conference series: materials science and engineering, vol
782. IOP Publishing, p 032008
Surya L (2017) Risk analysis model that uses machine learning to predict the likelihood of a fire occurring
at a given property. Int J Creat Res Thoughts (IJCRT), 2320–2882
Swiercz M, Mroczkowska H (2019) Application of PCA for early leak detection in a pipeline system of a
steam boiler. Prz. Elektrotechniczny Electr. Rev 95:190–203
Syed D, Abu-Rub H, Ghrayeb A, Refaat SS (2021) Household-level energy forecasting in smart buildings
using a novel hybrid deep learning model. IEEE Access 9:33498–33511
Tagliabue LC, Cecconi FR, Rinaldi S, Ciribini ALC (2021) Data driven indoor air quality prediction in edu-
cational facilities based on IoT network. Energy Build 236:110782
Taheri S, Ahmadi A, Mohammadi-Ivatloo B, Asadi S (2021) Fault detection diagnostic for HVAC systems
via deep learning algorithms. Energy Build 250:111275
Taheri S, Razban A (2021) Learning-based CO2 concentration prediction: application to indoor air quality
control using demand-controlled ventilation. Build Environ 108164
Tanabe S-I, Iwahashi Y, Tsushima S, Nishihara N (2013) Thermal comfort and productivity in offices under
mandatory electricity savings after the Great East Japan earthquake. Archit Sci Rev 56(1):4–13
Tang S, Shelden DR, Eastman CM, Pishdad-Bozorgi P, Gao X (2020) Bim assisted building automation
system information exchange using bacnet and ifc. Autom Constr 110:103049
Tariq S, Loy-Benitez J, Nam K, Lee G, Kim M, Park D, Yoo C (2021) Transfer learning driven sequential
forecasting and ventilation control of pm2.5 associated health risk levels in underground public facili-
ties. J Hazard Mater 406:124753
Taştan M, Gökozan H (2019) Real-time monitoring of indoor air quality with internet of things-based
e-nose. Appl Sci 9(16):3435
Tian Z, Si B, Shi X, Fang Z (2019) An application of Bayesian network approach for selecting energy effi-
cient HVAC systems. J Build Eng 25:100796

13
5018 Y. Himeur et al.

Tian Y, Yu J, Zhao A (2020) Predictive model of energy consumption for office building by using improved
GWO-BP. Energy Rep 6:620–627
Tien PW, Wei S, Calautit JK, Darkwa J, Wood C (2020) A vision-based deep learning approach for the
detection and prediction of occupancy heat emissions for demand-driven control solutions. Energy
Build 226:110386
Tien PW, Wei S, Calautit J (2021) A computer vision-based occupancy and equipment usage detection
approach for reducing building energy demand. Energies 14(1):156
Tiwari A, Batra U Blockchain enabled reparations in smart buildings-cyber physical system. Defence Sci J
71(4)
Tokarski M Protection of individuals in the light of eu regulation 2016/679 on the protection of natural
persons with regard to the processing of personal data and on the free movement of such data. Safety
Defense 2
Trianti-Stourna E, Spyropoulou K, Theofylaktos C, Droutsa K, Balaras C, Santamouris M, Asimakopou-
los D, Lazaropoulou G, Papanikolaou N (1998) Energy conservation strategies for sports centers:
part A. Sports halls. Energy Build 27(2):109–122
Valenzuela VEL, Lucena VF, Parvaresh P, Jazdi N, Göhner P (2013) Voice-activated system to remotely
control industrial and building automation systems using cloud computing. In: 2013 IEEE 18th
conference on emerging technologies & factory automation (ETFA). IEEE, pp 1–4
Valgaev O, Kupzog F (2017) Schmeck H (2017) Building power demand forecasting using k-nearest
neighbours model-practical application in smart city demo aspern project. CIRED-Open Access
Proc J 1:1601–1604
Valladares W, Galindo M, Gutiérrez J, Wu W-C, Liao K-K, Liao J-C, Lu K-C, Wang C-C (2019) Energy
optimization associated with thermal comfort and indoor air control via a deep reinforcement
learning algorithm. Build Environ 155:105–117
Van Cutsem O, Dac DH, Boudou P, Kayal M (2020) Cooperative energy management of a community of
smart-buildings: a blockchain approach. Int J Electric Power Energy Syst 117:105643
Van Engelen JE, Hoos HH (2020) A survey on semi-supervised learning. Mach Learn 109(2):373–440
Varlamis I, Sardianos C, Chronis C, Dimitrakopoulos G, Himeur Y, Alsalemi A, Bensaali F, Amira A
(2022) Smart fusion of sensor data and human feedback for personalized energy-saving recom-
mendations. Appl Energy 305:117775
Varlamis I, Sardianos C, Chronis C, Dimitrakopoulos G, Himeur Y, Alsalemi A, Bensaali F, Amira A
(2022) Using big data and federated learning for generating energy efficiency recommendations.
Int J Data Sci Anal, pp 1–17
Verma S, Singh S, Majumdar A (2019) Multi label restricted boltzmann machine for non-intrusive load
monitoring. In: ICASSP 2019–2019 IEEE international conference on acoustics, speech and signal
processing (ICASSP). IEEE, pp 8345–8349
Wang S (2020) Wireless network indoor positioning method using nonmetric multidimensional scaling
and rssi in the internet of things environment. Math Problems Eng 2020 8830891
Wang S (2009) Intelligent buildings and building automation. Routledge, London
Wang Z, Hong T (2020) Reinforcement learning for building controls: the opportunities and challenges.
Appl Energy 269:115036
Wang Y, Velswamy K, Huang B (2017) A long-short term memory recurrent neural network based rein-
forcement learning controller for office heating ventilation and air conditioning systems. Processes
5(3):46
Wang J, Li G, Chen H, Liu J, Guo Y, Sun S, Hu Y (2018) Energy consumption prediction for water-
source heat pump system using pattern recognition-based algorithms. Appl Therm Eng
136:755–766
Wang Y, Chen J, Chen X, Zeng X, Kong Y, Sun S, Guo Y, Liu Y (2020) Short-term load forecasting for
industrial customers based on tcn-lightgbm. IEEE Trans Power Syst 36(3):1984–1997
Wang R, Lu S, Feng W (2020) A novel improved model for building energy consumption prediction
based on model integration. Appl Energy 262:114561
Wang JQ, Du Y, Wang J (2020) LSTM based long-term energy consumption prediction with periodicity.
Energy 197:117197
Wang J, Chen Y (2021) Adaboost-based integration framework coupled two-stage feature extraction
with deep learning for multivariate exchange rate prediction. Neural Process Lett 1–25
Wang Z, Liu J, Zhang Y, Yuan H, Zhang R, Srinivasan RS (2021) Practical issues in implementing
machine-learning models for building energy efficiency: moving beyond obstacles. Renew Sustain
Energy Rev 143:110929

13
AI‑big data analytics for building automation and management… 5019

Wan B, Xu C, Mahapatra RP, Selvaraj P (2021) Understanding the cyber-physical system in interna-
tional stadiums for security in the network from cyber-attacks and adversaries using AI. Wireless
Personal Communications, pp 1–18
Wei W, Ramalho O, Malingre L, Sivanantham S, Little JC, Mandin C (2019) Machine learning and sta-
tistical models for predicting indoor air quality. Indoor Air 29(5):704–726
Wen L, Zhou K, Yang S (2020) Load demand forecasting of residential buildings using a deep learning
model. Electric Power Syst Res 179:106073
Wilcox HS (2020) Virtual metering for monitoring building energy consumption. Tech. rep., Los Ala-
mos National Lab.(LANL), Los Alamos, NM
Wu K, Wu J, Feng L, Yang B, Liang R, Yang S, Zhao R (2021) An attention-based CNN-LSTM-BiL-
STM model for short-term electric load forecasting in integrated energy system. Int Trans Electric
Energy Syst 31(1)
Wu L, Kong C, Hao X, Chen W (2020) A short-term load forecasting method based on gru-cnn hybrid
neural network model. Mathematical problems in engineering
Wu L, Wang Y Stationary and moving occupancy detection using the sleepir sensor module and machine
learning, IEEE Sens J
Wytock M, Kolter Z (2013) Sparse gaussian conditional random fields: algorithms, theory, and appli-
cation to energy forecasting. In: International conference on machine learning, PMLR, pp
1265–1273
Xiao Z, Gang W, Yuan J, Zhang Y, Fan C (2021) Cooling load disaggregation using a nilm method
based on random forest for smart buildings. Sustain Cities Soc 74:103202
Xiao-wei X (2020) Study on the intelligent system of sports culture centers by combining machine
learning with big data. Pers Ubiquit Comput 24(1):151–163
Xu C, Chen H (2020) A hybrid data mining approach for anomaly detection and evaluation in residential
buildings energy data. Energy Build 215:109864
Xu C, Wang J, Zhang J, Li X (2021) Anomaly detection of power consumption in yarn spinning using
transfer learning. Comput Ind Eng 152:107015
Yahyaoui A, Yaakoubi F, Abdellatif T et al (2020) Machine learning based rank attack detection for
smart hospital infrastructure. In: International conference on smart homes and health telematics.
Springer, pp 28–40
Yaïci W, Krishnamurthy K, Entchev E, Longo M (2021) Recent advances in internet of things (IoT)
infrastructures for building energy systems: A review. Sensors 21(6):2152
Yang S, Wan MP (2022) Machine-learning-based model predictive control with instantaneous line-
arization—a case study on an air-conditioning and mechanical ventilation system. Appl Energy
306:118041
Yang S, Wan MP, Chen W, Ng BF, Dubey S (2020) Model predictive control with adaptive machine-
learning-based model for building energy efficiency and comfort optimization. Appl Energy
271:115147
Yang S, Wan MP, Chen W, Ng BF, Dubey S (2021) Experiment study of machine-learning-based approx-
imate model predictive control for energy-efficient building control. Appl Energy 288:116648
Yoon S, Yu Y (2018) Strategies for virtual in-situ sensor calibration in building energy systems. Energy
Build 172:22–34
Yu Y (2011) Ai chiller: an open IoT cloud based machine learning framework for the energy saving of
building HVAC system via big data analytics on the fusion of bms and environmental data. arXiv:​
2011.​01047
Yu Y, Li H (2015) Virtual in-situ calibration method in building systems. Autom Constr 59:59–67
Yu Z, Haghighat F, Fung BC, Yoshino H (2010) A decision tree method for building energy demand
modeling. Energy Build 42(10):1637–1646
Yu J, Jang J, Yoo J, Park JH, Kim S (2017) Bagged auto-associative kernel regression-based fault detec-
tion and identification approach for steam boilers in thermal power plants. J Electr Eng Technol
12(4):1406–1416
Yuan Z, Wang W, Wang H, Mizzi S (2020) Combination of cuckoo search and wavelet neural network
for midterm building energy forecast. Energy 202:117728
Yucong W, Bo W (2020) Research on ea-xgboost hybrid model for building energy prediction. J Phys
1518:012082
Yun W-S, Hong W-H, Seo H (2021) A data-driven fault detection and diagnosis scheme for air handling
units in building HVAC systems considering undefined states. J Build Eng 35:102111
Yu L, Qin S, Zhang M, Shen C, Jiang T, Guan X. A review of deep reinforcement learning for smart
building energy management. IEEE Internet Things J 8(1)5 12046–12063

13
5020 Y. Himeur et al.

Zakharchenko A, Stepanets O (2019) Edge computing in building automation system-pros and cons. In:
Modeling, control and information technologies: proceedings of international scientific and practi-
cal conference, pp 130–132
Zekić-Sušac M, Has A, Knežević M (2021) Predicting energy cost of public buildings by artificial neural
networks, cart, and random forest. Neurocomputing 439:223–233
Zhan S, Liu Z, Chong A, Yan D (2020) Building categorization revisited: a clustering-based approach to
using smart meter data for building energy benchmarking. Appl Energy 269:114920
Zhang Q, Fu F, Tian R (2020) A deep learning and image-based model for air quality estimation. Sci
Total Environ 724:138178
Zhang C, Li J, Zhao Y, Li T, Chen Q, Zhang X (2020) A hybrid deep learning-based method for short-term
building energy load prediction combined with an interpretation process. Energy Build 225:110301
Zhang G, Tian C, Li C, Zhang JJ, Zuo W (2020) Accurate forecasting of building energy consumption
via a novel ensembled deep learning method considering the cyclic feature. Energy 201:117531
Zhang G, Li Y, Deng X (2020) K-means clustering-based electrical equipment identification for smart
building application. Information 11(1):27
Zhang L, Leach M, Bae Y, Cui B, Bhattacharya S, Lee S, Im P, Adetola V, Vrabie D, Kuruganti T (2021)
Sensor impact evaluation and verification for fault detection and diagnostics in building energy sys-
tems: a review. Adv Appl Energy 3(1):100055
Zhang Y, Geng P, Sivaparthipan C, Muthu BA (2021) Big data and artificial intelligence based early risk
warning system of fire hazard for smart cities. Sustain Energy Technol Assess 45:100986
Zhang L, Wen J, Li Y, Chen J, Ye Y, Fu Y, Livingood W (2021) A review of machine learning in building
load prediction. Appl Energy 285:116452
Zhang X, Zeng Z, Wang P, Song J, Kong Z (2020) A hybrid edge-cloud computing method for short-term
electric load forecasting based on smart metering terminal. In: 2020 IEEE 4th conference on energy
internet and energy system integration (EI2). IEEE, pp 3101–3105
Zhao H, Hua Q, Chen H-B, Ye Y, Wang H, Tan SX-D, Tlelo-Cuautle E (2018) Thermal-sensor-based occu-
pancy detection for smart buildings using machine-learning methods. ACM Trans Design Autom
Electr Syst (TODAES) 23(4):1–21
Zhao Y, Zhang C, Zhang Y, Wang Z, Li J (2020) A review of data mining technologies in building energy
systems: load prediction, pattern identification, fault detection and diagnosis. Energy Built Environ
1(2):149–164
Zheng H, Yuan J, Chen L (2017) Short-term load forecasting using EMD-LSTM neural networks with a
xgboost algorithm for feature importance evaluation. Energies 10(8):1168
Zheng P, Wang C, Liu Y, Lin B, Wu H, Huang Y, Zhou X (2022) Thermal adaptive behavior and ther-
mal comfort for occupants in multi-person offices with air-conditioning systems. Build Environ
207:108432
Zhong H, Wang J, Jia H, Mu Y, Lv S (2019) Vector field-based support vector regression for building
energy consumption prediction. Appl Energy 242:403–414
Zhong F, Calautit JK, Hughes BR (2020) Analysis of the influence of cooling jets on the wind and thermal
environment in football stadiums in hot climates. Build Serv Eng Res Technol 41(5):561–585
Zhou K, Yang S (2016) Understanding household energy consumption behavior: the contribution of energy
big data analytics. Renew Sustain Energy Rev 56:810–819
Zhu X, Chen K, Anduv B, Jin X, Du Z (2021) Transfer learning based methodology for migration and
application of fault detection and diagnosis between building chillers for improving energy efficiency.
Build Environ 200:107957
Zou J, Zhao Q, Yang W, Wang F (2017) Occupancy detection in the office by analyzing surveillance videos
and its application to building energy conservation. Energy Build 152:385–398
Zubaidi SL, Gharghan SK, Dooley J, Alkhaddar RM, Abdellatif M (2018) Short-term urban water demand
prediction considering weather factors. Water Resour Manage 32(14):4527–4542
Zverovich V, Mahdjoubi L, Boguslawski P, Fadli F, Barki H (2016) Emergency response in complex
buildings: automated selection of safest and balanced routes. Comput Aided Civil Infrastr Eng
31(8):617–632

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

13
AI‑big data analytics for building automation and management… 5021

Authors and Affiliations

Yassine Himeur1,2 · Mariam Elnour1 · Fodil Fadli1 · Nader Meskin3 · Ioan Petri4 ·
Yacine Rezgui4 · Faycal Bensaali3 · Abbes Amira5,6
Mariam Elnour
[email protected]
Fodil Fadli
[email protected]
Nader Meskin
[email protected]
Ioan Petri
[email protected]
Yacine Rezgui
[email protected]
Faycal Bensaali
[email protected]
Abbes Amira
[email protected]; [email protected]
1
Department of Architecture & Urban Planning, Qatar University, Doha, Qatar
2
College of Engineering and Information Technology, University of Dubai, Dubai, UAE
3
Department of Electrical Engineering, Qatar University, Doha, Qatar
4
School of Engineering, BRE Institute of Sustainable Engineering, Cardiff University, Wales, UK
5
Department of Computer Science, University of Sharjah, Sharjah, UAE
6
Institute of Artificial Intelligence, De Montfort University, Leicester, UK

13