Hindawi

Mobile Information Systems

Volume 2022, Article ID 6181357, 9 pages
https://doi.org/10.1155/2022/6181357

Research Article
Convolutional Neural Network Based Energy Consumption
Management Model for the Full Life Cycle of Buildings and
Information System Design

Jingyi Zhou
Department of Management Engineering, Hunan Urban Construction College, Xiangtan 411100, China

Correspondence should be addressed to Jingyi Zhou; [email protected]

Received 29 June 2022; Revised 26 July 2022; Accepted 29 July 2022; Published 15 September 2022

Academic Editor: Santosh Tirunagari

Copyright © 2022 Jingyi Zhou. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
With the continuous improvement in China’s economy, the construction industry has developed and rampantly progressed.
Besides the wastage of resources and energy, the development has caused serious pollution to the environment. This makes the
construction industry a high energy-consuming and highly polluting industry. There is a pressing need to reduce the wastage of
resources and to adequately manage consumption of energy throughout the life cycle of buildings. This paper explores an effective
method of building life cycle energy management by appropriately utilizing information system and the emerging deep learning
technology. To achieve energy saving in buildings, a feasible model is proposed for predicting, analyzing, and building energy
consumption based on neural networks. By analyzing the massive data stored in the building information system, the operation of
each subsystem in the building is guided and regulated to achieve energy deployment and build energy optimization. Focusing the
key meters, the average generalization ability of the proposed model (R-Squared  1.9, MSE  1.02) is better than the other
contemporarily used models, LightGBM, LSTM, and SVR. Moreover, the method can effectively predict the energy consumption
of the whole life cycle of the building and has higher prediction accuracy. The method proposed has great significance in research
related with improving building energy performance and designing decision support tool.

1. Introduction building energy management to multiple stages, such as

project, design, construction and operation, and covering
In recent years, with strong advocacy and active promotion the whole life cycle of the building, can achieve the goal of
of the Ministry of Housing and Construction and local controlling and saving energy. The goal is put into practice
building energy efficiency authorities, constructions of large through the promotion of energy-saving indicators and
public buildings are increasing adamantly [1–3]. Buildings technical measures at each stage. Effective management of
are the major energy consumers, and many buildings in energy consumption during the whole life cycle of a building
China are high energy-consuming buildings, which impose a is just getting started in China. The information on energy
heavy energy burden on society [4–6]. The annual energy consumption during each construction phase is distributed
consumption of public and civil buildings is also increasing among different departments for apt compliance. However,
with the passage of time. According to a recent research [7], the collection of relevant data and the standards of infor-
the energy consumed by buildings accounts for 40% of the mation system construction are yet to be established.
whole energy consumption, causing a major chunk of at- At present, people mainly focus on the improvement and
mospheric emissions (40%) and greenhouse gas emission optimization of energy consumption during building use,
(33%). including energy consumption in heating, air conditioning,
However, building energy management is a systematic lighting, household appliances, and cooking utensils. In fact,
project, and the actual energy consumption after the each stage of the building life cycle, from design to con-
building is put into operation may be reduced. By extending struction completion, delivery, and use until demolition, is
2 Mobile Information Systems

constantly consuming energy. Moreover, with the contin- In 2012, in the deep learning (DL) technology, con-
uous emergence of new materials, technologies, and tech- volutional neural network (CNN) emerged in the field of
niques, the proportion of energy consumed in the building image recognition that attracted the attention of research
materials and construction process is relatively higher. scholars all over the world. The CNN technology has a wide
Therefore, it is not enough to consider only the energy range of applications in the field of computer vision, natural
consumption in the operation phase of a building but the language processing, and other research fields. Compared
energy consumption in the whole life cycle of a building with ordinary neural network technology, CNN has pow-
should be systematically considered. According to the life erful data feature extraction ability with convolutional layers
cycle theory of buildings, the life cycle of a building includes and has better generalization ability. The model proposed in
five stages: the preparation stage of building materials, the this paper for the whole-life energy consumption manage-
building construction stage, the building use stage, the ment is based on CNN. The aim behind the research study is
building demolition stage, and the disposal stage of used to provide preliminary findings and relevant technology
building materials, see Figure 1. preparation for further research and practice in the field of
Systems engineering is a dedicated domain which deals construction. To realize the whole-life energy consumption
with materials, construction techniques, and other factors of management of buildings, firstly, it is necessary to determine
construction. Systems engineering requires a system per- the monitoring target, i.e., to split the comprehensive control
spective, focuses on the interconnectedness, interaction, and target of the whole-life energy consumption of buildings into
mutual constraints between the whole and its parts, and several implementable, measurable, and controllable indi-
between the whole and the outside [8–12]. The openness, cators, to choose a suitable management stage, and then to
relevance, and dynamics of building systems determine that refine and decompose each of the indicators. Secondly, it is
energy saving in building systems must consider the whole necessary to determine the implementation path, integrate
process from energy and resource acquisition to transmis- requirements of energy consumption, and design a suitable
sion and distribution. For example, double-glazed windows energy consumption management process.
are better insulated than ordinary wood windows and doors, In this paper, we sort out the logical relationships among
but double-glazed windows consume more energy in the the monitoring objects, implementation paths, and expected
production process than the production of ordinary wood results of energy consumption in the whole life cycle. A
windows and doors. Therefore, when choosing building model for energy consumption management for the whole
solutions, the energy consumption of the whole life cycle life cycle of buildings is proposed to provide guidance for the
should be considered from a system perspective. development of information systems. The model con-
Building information system is a digital technology to struction schematic is shown in Figure 2.
simulate the real information of the building by con-
structing digital information system through information 2. Related Work
technology. The information description is to clearly and
intuitively obtain the energy consumption in different Building construction is a compound process encompassing
seasons and to provide guidance for building construction various phases and plans [18]. Energy consumption control
and environmental protection. The information system requirements in the preproject stage and indexes in the
provides guidance for building construction and envi- project planning stage are used as the baselines for the design
ronmental protection, thus controlling energy and resource stage. The data linkage with the building energy con-
waste and saving construction costs, which is important for sumption control requirements is established during the
the sustainable development of the construction industry project planning phase. Moreover, analysis about invest-
[13]. Moreover, the technology may be used to dynamically ment, sales, social benefit analysis, risk, project investment,
reflect building information in real-time. The application of and financing are also performed in the planning phase. In
building information system for the whole life cycle of the design phase, dedicated software technology, compre-
building focuses on the continuity of information. The hensive database, and system model are used to improve the
whole life cycle of building contains four stages such as efficiency of related building energy consumption, green
planning and design, construction, operation, and main- building, and other analysis tools and also to reduce the huge
tenance, and it is very important to keep the continuity and workload during the whole process. Moreover, the building
connection of the four stages [14]. The building informa- information system is used to achieve dynamic output of
tion system contains a lot of data and information related to building energy analysis such as electromechanics, archi-
building performance, so the building information system tecture, Heating, Ventilation, and Air Conditioning
software needs to export the relevant data first and then (HVAC), and curtain wall and to provide a basis for
import them into the building information system database parametric design. Quantitative indicators in the informa-
for integration and analysis after conversion. Following tion system like system indicators, equipment parameters,
that, the indicators are integrated and adjusted to control and operation parameters are helpful to integrate and adjust
energy requirements of the building. The data and the energy consumption of a building [19–24].
parameters related to energy consumption are imported The construction phase is an important stage where
into the comprehensive database of the building infor- design of the building energy management is meticulously
mation system to form quantitative index parameters brought into actuality. Integration and adjustment may be
[15–17]. performed in the comprehensive database of the building
Mobile Information Systems 3

Energy consumption of construction raw materials,

intermediate materials and finished and semi-finished products
Building material
for mining, production and processing and the transportation
preparation stage
process involved in the intermediate links in the construction
and use phase of the building for daily maintenance and repair

The energy consumption of materials, machinery processing,

Building construction
construction consumption at the site and all transportation
stage
processes involved in the intermediate stages.

Energy consumption of heating, air conditioning, ventilation,

Building use phase hot water supply, lighting, cooking, household appliances,
elevators, etc. in the daily use of the building.

Building demolition The consumption required when the building is dismantled and
stage demolished.

The energy consumption involved in the transportation, sorting,

treatment, reuse and disposal of useless waste building
Waste building materials
materials in the final disposal site during the daily maintenance
disposal stage
phase of the building and after demolition of the waste
building materials generated

Figure 1: Energy consumption in all phases of the building life cycle.

Full Life Cycle Energy Management Model for Buildings

Monitoring Implement
Objects ation Path
Business
Expected
Information
Results
Indicator 1 Design management process System
Indicator 2 Identify responsible subjects
... Selecting management stages
Indicator N

Figure 2: Schematic of the building life cycle energy management models.

information system. However, special attention is paid to building energy management in conjunction with other
include the comparison of equipment parameters binding information about building operation. On the basis of real-
indexes with actual procurement according to the com- time collection of dynamic data information such as human
parison, actual installation, piping, and construction flow, environment, and operation of facilities and equip-
drawings. At the same time, since a project may undergo ment, it integrates real-time data and historical data of
relevant changes due to uncontrollable factors such as various types of energy consumption in the building, ex-
equipment manufacturers, construction environment, and tracts relevant information from the building information
building layout, the energy consumption and control pa- system model. Data simulation and analysis techniques may
rameters are kept adjustable [25–28]. The key inputs (or be used to carry out simulation estimation of operational
constraints) in the construction phase include various pa- energy consumption under various conditions. After the
rameters of electromechanical equipment (performance, building operation is stabilized (generally two heating and
equipment life, and index), installation location, supplier cooling cycles), the building energy management system
information, operation monitoring and control index, the collects dynamic data such as the optimal performance curve
correlation of electromechanical systems (piping diagram, of equipment operation, the optimal life curve of equipment
logic of the cooling, heat source system and air conditioning operation, and the monitoring data of the operating
system), variable air volume (VAV) system, and area space equipment [26].
layout. With the development of information technology, a huge
The operation phase is the phase to test the results of amount of complicated information is flooded around all the
building energy management and to adjust and optimize time. However, the information that human beings can
4 Mobile Information Systems

receive is limited, and researchers have discovered that the represented by GFS and BigTable. The main representative of
human visual system has a huge visual information pro- master-slave is Hadoop. The later storage architecture is de-
cessing capacity with a limited field of view [27–29]. veloped by Amazon whereas its main representative is Dynamo.
Mimicking the human visual system, researchers have de- The Hadoop master-slave architecture is mainly reflected in the
veloped the idea of attentional mechanisms. The core idea of architecture design of HDFS (Hadoop Distributed File System).
the attention mechanism is to obtain the importance of each
feature map by certain means and devote more computa-
tional resources of the neural network to more important 3.3. Feature Embedding. Word embedding refers to a kind of
tasks and use the results of the tasks to guide the update of word representation where words with similar meaning have a
weights of the feature maps so that the corresponding tasks similar representation. Word embedding is normally used in
can be completed efficiently and quickly [30]. natural language processing. The embedding layer maps sparse
Convolutional neural network (CNN) is a widely used word vectors to a low-dimensional and compact feature space.
model in deep learning, which is an important part of target The vectors in this feature space can be used to measure the
detection algorithm and plays the role of feature extractor in similarity between features by computing their relative distance.
target detection algorithm mainly completing the feature ex- Since the feature space is a feature compression of the original
traction task and outputting the feature map containing rich vector space, its dimensionality is much smaller than that of the
feature information, which lays the foundation for the sub- original vector space. Hence, the complexity of distance cal-
sequent classification and regression tasks in target detection. culation can be greatly reduced. Moreover, the undesirable
In 2012, the AlexNet network was proposed to focus research in effects of expanded feature vectors are avoided.
the field of computer vision on convolutional neural networks In the N-dimensional word vector space VERX, the
and deep learning. With the emergence of various advanced relationship between its features can be expressed in terms of
frameworks in 2014, improved AlexNet and ZFNet were in- conditional probabilities as
troduced [31–33]. In the same year, RepVGG, which consists of N
only 3 × 3 convolutions with ReLU activation functions further p v1 , v2 , · · · , vn ∣ vi 􏼁 � 􏽙 p􏼐vj ∣ vi 􏼑, (1)
enhances the feature extraction performance of the VGG j≠i
network through a simple branchless structure [34–36]. To where vi is the target feature and vj is the other features in the
realize the deployment of CNN models in miniaturized mobile feature space. In the neural network, p(vj ∣ vi ) can be
and embedded platforms, scholars have also conducted in- expressed as
depth research on how to reduce the number of CNN model exp sj
parameters and the complexity of CNN structures. Among p 􏼐v j ∣ v i 􏼑 � M , (2)
􏽐j′ �1 exp sj′
them, SqueezeNet utilizes many 1 × 1 convolutional kernels to
replace 3 × 3 convolutional kernels while reducing the number where H is the hidden layer, M is the dimension of the
of channels of 3 × 3 convolutional kernels to reduce the original N-dimensional feature mapping to the feature space
number of parameters. S, sj is the j-th component of the weight W of the hidden
layer H to the feature mapping space S, and h and W is the
3. Methods weight of the original feature input to the hidden layer H.
Therefore, given the target feature vi , the loss function of the
Machine learning models work as a black box discovering
network can be obtained as
the relation between various features of building and gen-
erate outputs about the energy performance. In this paper, N

the supervised ML approach used adequate data about L � − 􏽘 ln p v1 , v2 , · · · , vN ∣ vi 􏼁

i�1
building to predict targets for unseen samples.
exp sj
� −ln 􏽙 (3)
3.1. Model Architecture. As any ML model works on data, 􏽐M
j′ �1 exp sj′
initially, the historical number of buildings and meteoro-
logical data are collected to have a dataset for the model N M
training. The algorithm is used to identify the building run � − 􏽘 sj + N ln 􏽘 expj′ .
cycles as the time steps of the model. Finally, the hybrid j�1 j′ �1
model is trained based on the convolutional neural network
algorithm. The optimal hyperparameters of the hybrid As clear from the equation, after the word embedding
model are found using the grid search algorithm. The layer, the relevance of the classification features can be fully
structure of the proposed model is shown in Figure 3. extracted, and the new features are embedded in a layer of
network. This reduces the dimensionality of the input data,
and the layer may be used for solving the problem of feature
3.2. Big Data Information System. With the continuous de- discretization. There are categorical features in building
velopment and wide application of big data technology, the metainformation, such as features related with the use and
platform architecture of big data is extending. The architecture location of buildings. Though the features have obvious
is divided into two basic types: (1) master-slave and (2) P2P. The effects on the energy consumption of the building, they do
former storage architecture is developed by Google and not exist in numerical form. If the categorical features are
Mobile Information Systems 5

Building Energy
Monitoring

CNN

Prediction
Category

Softmax

Figure 3: Structure of the proposed model.

coded with the unique thermal coding, it will lead to too split. The output of each cardinal module, V, represents the
large data dimensionality. For this reason, the fusion neural output of cardinal with channel weights, a(c) is the weight
network established in this study introduces an embedding calculated by SoftMax, and G represents the weight of each
layer, which is used to compute the embedding of discrete split. The final k outputs are stitched after the cardinal
classification signals, thus mapping the discrete classification module combine the information of the k cardinal outputs
features to a continuous word embedding space enabling the and the stitched outputs are element-wise summation with
fusion of discrete features with numerical features. At the the original inputs to obtain the final output.
same time, the word embedding layer reduces the data Rk
dimensionality and reduces the training time of the network. 􏽢k �
U 􏽘 Ui ,
i�R(k−1)+1

R
3.4. Convolutional Layer. The convolutional neural network
Vkc � 􏽘 aki (c)UR(k−1)+i ,
is selected to perform regression on the data. Convolution is
i�1
a mathematical operation in which the process is to take a
tensor, matrix, or vector and pass it through the convolution ⎪
⎧
⎨ exp 􏼐Gci 􏼐sk 􏼑􏼑 (4)
operation of a convolution kernel to obtain a tensor of aki (c) � ⎪ R , R > 1,
smaller dimension containing feature information. Deep ⎩ 􏽐 G c 􏼐 sk 􏼑
j�1 j
convolutional neural networks based on two-dimensional
convolutional kernels have made significant breakthroughs 1
in image recognition in recent years. However, the two- , R � 1,
1 + exp 􏼐−Gci 􏼐sk 􏼑􏼑
dimensional convolution kernel operates on two dimensions
of image data, namely, length and width. Considering that
the dimension of time-series data is only one-dimensional, V � Concat􏽮V1 , V2 , · · · , Vk 􏽯.
this study selects one-dimensional convolution to extract
time-series features from the data. The one-dimensional
convolution kernel convolves the time-series data on the 3.5. Model Training. As a first step, preprocessing of various
time series. types of data involved in building energy consumption is
By inputting the input of the network into k branches, performed. Particularly data dissimilar in scales are pre-
the input of each cardinal, where R represents the number of processed to avoid computational cost. There are various
branches after split in each cardinal, k represents the k-th types of data such as pressure, temperature, voltage, current,
cardinal, and U represents the input of each branch after and flow. Abnormalities of various forms may occur in the
6 Mobile Information Systems

data due to many reasons, including change in the envi- Table 1: The model parameters setting.
ronment. Therefore, it is needed to normalize data so that it Predictive models Parameter name Parameters
is not affected by various types of magnitudes. In this paper,
Number of hidden layer neurons 100
the maximum-minimum normalization method is used to Dropout 0.2
reduce abnormalities in data. CNN
Learning rate 0.001
x − Amin Optimization algorithm Adam
xnom � . (5)
Amax − Amin
The mean absolute percentage error (MAPE) is used as The training process performance enhancement and loss
the evaluation criterion for the outcome error as follows. convergence are shown in Figures 4 and 5.
N 􏼌􏼌􏼌 􏼌
􏼌y􏽢 − yi 􏼌􏼌􏼌􏼌 100
MAPE � 􏽘 􏼌􏼌􏼌 i 􏼌 , (6)
􏼌 y 􏼌􏼌 N
i�1 i
4.2. Experimental Results and Analysis. To evaluate the
prediction performance of the proposed models, LightGBM,
where N is the total number of prediction experiments, yi is CatBoost, support vector machine regression, and long
􏽢 i is the predicted value. The smaller the value
the true value, y short-term memory (LSTM) network were selected as the
of MAPE, the smaller the difference between the predicted cross-sectional comparison models. LightGBM and Cat-
value and the true value. Boost algorithms are both improvements of the gradient
boosted decision tree (GBDT) algorithm. Compared to the
4. Experiments and Results GBDT algorithm, which slices features at each level, the
LightGBM algorithm slices features directly at the leaf nodes
To design a suitable energy consumption management of the tree and introduces histogram optimization without
process, it is necessary to integrate requirements of energy sorting each feature, resulting in a significant speedup
consumption, after preprocessing and appropriate analysis of compared to the GBDT algorithm. Both the integrated
enough volume of data. Therefore, the CNN-based model learning algorithms are widely used in the field of data
proposed in this paper is trained by a dataset of 800,000 mining. Meanwhile, in order to verify that the one-di-
samples. mensional convolutional neural network (Conv1D) has the
performance of extracting temporal features in temporal
order, a long- and short-term memory network is selected
4.1. Experimental Setup. The experimental configuration for comparison.
environment is Ubuntu 18.04LTS with 32G RAM, Intel From the results in Figures 6, 7, and 8, the model
Corei7-7700 CPU and Nvidia GTX-1070Ti GPU, and Keras proposed in this study possesses high performance in terms
deep learning framework and TensorFlow as Backend. The of accuracy and model interpretation, which is only slightly
data used in this study come from the internal nonpublic lower than the LightGBM algorithm and better than other
data of a Chinese construction company, and the historical integrated learning algorithms and neural networks. The
data of different buildings in different places for one year are model in this study consumes more time in convergence,
collected. The main information contained in the data in- which is caused by the search of hyperparameters for more
clude meter (electricity meter, cooling meter, steam meter, iterations of training. Compared with the LSTM model, the
and heat meter) readings, building metainformation (e.g., LIGHTGBM model has higher prediction accuracy, and
usage, commissioning time, number of floors, building lo- compared with SVR and GA algorithms, TSA has better
cation) and meteorological information (e.g., air tempera- merit-seeking ability and convergence in optimizing the
ture, cloud cover, dew point temperature, air pressure, and LSTM model, indicating that the TSA algorithm is suitable
wind speed). Correspondingly, there are test datasets in the for parameter optimization of the LSTM model. Compared
data that have the same structure as the training dataset. For with the three single neural network models LSTM,
the training data selection, the original data, after the LIGHTGBM, and LSTM, the LIGHTGBM-LSTM model has
missing value processing, have 11714696 items, containing better prediction accuracy and robustness and has stronger
27 temporal features and 3 classification features. This study generalization ability. This indicates that the hybrid pre-
divided the data into training set and test set according to the diction model proposed makes full use of the advantages of
ratio of 4 : 1; however, in the process of training the model, different neural networks and has better prediction per-
the author found that the training speed of SVR and LSTM formance. As the model meets the practical engineering
on large datasets was too slow; therefore, for the above two needs, it provides effective data support for the power system
models, this study only selected 800,000 of these samples as of buildings.
the dataset for training the model and also divided the In order to verify the generalization ability of the model,
training set and test set according to the ratio of 4 : 1. In this regression prediction is performed for each of the other
study, the early stopping strategy is used to stop the training three measures in the paper, and the LightGBM algorithm is
to prevent overfitting of the models. Also, k-fold cross- used to compare with the model proposed in this study. In
validation is used for all algorithms, and the k-value is set to the LightGBM algorithm, the parameters are set in the same
6, and the results are averaged over six times. The model way as those for the meter prediction. For the parameters of
parameters were set as shown in Table 1. the model in this study, the parameters of the electricity
Mobile Information Systems 7

120

100

Loss 80

60

40

20

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
epoch

Training set
Validation set
Figure 4: Schematic diagram of training process performance improvement.

120

100

80
Fitting

60

40

20

0
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39
epoch

Training set
Validation set
Figure 5: The training process loss convergence schematic.

MSE R-Square
This study model + TPE This study model + TPE
This study model+GS This study model+GS
This model This model
SVR SVR
LSTM LSTM
XGBoost XGBoost
CatBoost CatBoost
LightGBM LightGBM
0 0.05 0.1 0.15 0.2 0.25 0.3 0.75 0.8 0.85 0.9 0.95 1
Figure 6: MSE comparison results. Figure 7: R-square comparison results.
8 Mobile Information Systems

Convergence time (min) management in the initial operation stage of different

This study model + TPE buildings. In short, the application of building information
This study model+GS
system is clearly and intuitively analyzed in terms of the
building energy consumption of the whole building cycle
This model
and provides a basis for the development of scientific energy
SVR
consumption management schemes. The efficient general-
LSTM
ization ability of the proposed model suits it well in the
XGBoost designing of decision support tools and to improve building
CatBoost energy performance. With the advent of Big Data and In-
LightGBM ternet of Things, advance sensors and energy meters are
0 10 20 30 40 50 60 70 80 90 100 required in buildings. In future, the method will be enhanced
to support the futuristic sensors and meters so as to meet the
Figure 8: Convergence time comparison results.
requirements of the upcoming data processing systems.

Table 2: R-square validation of model generalization capability.

Data Availability
MSE
Meters The datasets used during the current study are available from
LightGBM Proposed model
the corresponding author on reasonable request.
Cooling meters 0.2467 0.216
Steam meters 0.6171 0.606
Heat meters 0.5800 0.600 Conflicts of Interest
The authors declare that they have no conflicts of interest.
Table 3: Model generalization capability MSE validation.
R-square References
Meters
LightGBM Proposed model [1] S. Ding, S. Qu, Y. Xi, and S. Wan, “A long video caption
Cooling meters 0.859 0.878 generation algorithm for big video data retrieval,” Future
Steam meters 0.833 0.840 Generation Computer Systems, vol. 93, pp. 583–595, 2019.
Heat meters 0.828 0.825 [2] Z. Gao, Y. Li, and S. Wan, “Exploring deep learning for view-
based 3D model retrieval,” ACM Transactions on Multimedia
Computing, Communications, and Applications, vol. 16, no. 1,
consumption prediction model were still used as the basis pp. 1–21, 2020.
for parameter setting, and the TPE algorithm was not used [3] Z. Gao, H. Xue, and S. Wan, “Multiple Discrimination and
for the superparameter search. The experimental results are Pairwise CNN for view-based 3D object retrieval,” Neural
shown in Tables 2 and 3. As seen in Tables 2 and 3, the Networks, vol. 125, pp. 290–302, 2020.
[4] S. Ding, S. Qu, Y. Xi, and S. Wan, “Stimulus-driven and
performance of the model proposed in this study is close to concept-driven analysis for image caption generation,”
or even better than that of the LightGBM algorithm. Neurocomputing, vol. 398, pp. 520–530, 2020.
It can be seen that the model proposed in this study has [5] J. Pei, Z. Yu, J. Li, M. A. Jan, and K. Lakshmanna, “TKAGFL: a
good generalization ability and is not only suitable for federated communication framework under data heteroge-
prediction of building electricity consumption but also can neity,” IEEE Transactions on Network Science and Engineering,
get good results when predicting energy consumption such vol. 1, 2022.
as building cooling and heat. [6] M. Zhao, C. Chen, L. Liu, D. Lan, and S. Wan, “Orbital
collaborative learning in 6G space-air-ground integrated
networks,” Neurocomputing, vol. 497, pp. 94–109, 2022.
5. Conclusion [7] A. Kosonen and A. Keskisaari, “Zero-energy log house
– future concept for an energy efficient building in the Nordic
With this research work, the building information system conditions,” Energy and Buildings, vol. 228, Article ID 110449,
based on convolutional neural network is proposed to an- 2020.
alyze building energy consumption of the four cycles- [8] A. M. Obeso, J. Benois-Pineau, A. A. R. Acosta, and
planning, design, construction, and operation-cum-main- M. S. G. Vazquez, “Architectural style classification of
tenance. The key elements of building construction with the Mexican historical buildings using deep convolutional neural
findings observed are summarized. Information integrity, networks and sparse features,” Journal of Electronic Imaging,
significance, and correlation of the building information vol. 26, no. 1, Article ID 011016, 2016.
system are discussed. In the proposed model, regardless of [9] X. Yang, H. Li, Y. Yu, X. Luo, T. Huang, and X. Yang,
“Automatic pixel-level crack detection and measurement
the quantitative data and correlation rules, features are
using fully convolutional network,” Computer-Aided Civil
passed through the comprehensive database of the building and Infrastructure Engineering, vol. 33, no. 12, pp. 1090–1109,
information system after digitization and quantification. The 2018.
building information system can simulate and predict the [10] H. L. Yang, J. Yuan, D. Lunga, M. Laverdiere, A. Rose, and
operation energy consumption in the visualized parameter B. Bhaduri, “Building extraction at scale using convolutional
environment and provide guidance for energy consumption neural network: mapping of the United States,” Ieee Journal of
Mobile Information Systems 9

Selected Topics in Applied Earth Observations and Remote [26] S. Kloepple, “Lighting upgrade breathes new life into data
Sensing, vol. 11, no. 8, pp. 2600–2614, 2018. center[J],” Buildings, vol. 112, no. 11, p. 10, 2018.
[11] J. Salamon and J. P. Bello, “Deep convolutional neural net- [27] D. Chakraborty and H. Elzarka, “Early detection of faults in
works and data augmentation for environmental sound HVAC systems using an XGBoost model with a dynamic
classification[J],” IEEE Signal Processing Letters, vol. 3, p. 1, threshold,” Energy and Buildings, vol. 185, no. FEB,
2017. pp. 326–344, 2019.
[12] Z. Guo, Q. Chen, G. Wu, Y. Xu, R. Shibasaki, and X. Shao, [28] R. E. Amaral, J. Brito, M. Buckman et al., “Waste management
“Village building identification based on ensemble convolu- and operational energy for sustainable buildings: a review,”
tional neural networks,” Sensors, vol. 17, no. 11, p. 2487, 2017. Sustainability, vol. 12, no. 13, p. 5337, 2020.
[13] E. Maltezos, N. Doulamis, A. Doulamis, and C. Ioannidis, [29] S. Wang and W. Gang, “Design and control optimization of
“Deep convolutional neural networks for building extraction energy systems of smart buildings today and in the near future
from orthoimages and dense image matching point clouds [J],” Frontiers of Engineering Management, vol. 4, no. 01,
[J],” Journal of Applied Remote Sensing, vol. 11, no. 4, pp. 62–70, 2017.
pp. 042620-042621, 2017. [30] C. Korkas, A. Dimara, I. Michailidis et al., “Integration and
[14] L. De Baets, J. Ruyssinck, C. Develder, T. Dhaene, and verification of PLUG-N-harvest ICT platform for intelligent
D. Deschrijver, “Appliance classification using VI trajectories management of buildings[J],” Energies, vol. 15, 2022.
and convolutional neural networks,” Energy and Buildings, [31] R. Sundling, A. Blomsterberg, and A. Landin, “Enabling
vol. 158, no. 1, pp. 32–36, 2018. energy-efficient renovation: the case of vertical extension to
[15] R. Alshehhi, P. R. Marpu, W. L. Woon, and M. D. Mura, buildings,” Construction Innovation, vol. 19, no. 1, pp. 2–14,
“Simultaneous extraction of roads and buildings in remote 2019.
sensing imagery with convolutional neural networks,” ISPRS [32] E. Corry, P. Pauwels, S. Hu, M. Keane, and J. O’Donnell, “A
Journal of Photogrammetry and Remote Sensing, vol. 130, performance assessment ontology for the environmental and
no. aug, pp. 139–149, 2017. energy management of buildings,” Automation in Construc-
[16] F. R. Cecconi, M. Manfren, L. C. Tagliabue, A. L. C. Ciribini, tion, vol. 57, pp. 249–259, 2015.
and E. De Angelis, “Probabilistic behavioral modeling in [33] R. Ali, M. H. Siddiqi, M. Idris et al., “GUDM: automatic
building performance simulation: a Monte Carlo approach,” generation of unified datasets for learning and reasoning in
Energy and Buildings, vol. 148, no. aug, pp. 128–141, 2017. healthcare,” Sensors, vol. 15, no. 7, pp. 15772–15798, 2015 Jul
[17] M. Potkany, M. Vetrakova, and M. Babiakova, “Facility 2.
management and its importance in the analysis of building life [34] M. Manic, D. Wijayasekara, K. Amarasinghe, and
cycle,” Procedia Economics and Finance, vol. 26, pp. 202–208, JJ. Rodriguez-Andina, “Building energy management systems:
2015. the age of intelligent and adaptive buildings,” IEEE Industrial
[18] C. Llatas, B. Soust-Verdaguer, and A. Passer, “Implementing Electronics Magazine, vol. 10, no. 1, pp. 25–39, 2016.
life cycle sustainability assessment during design stages in [35] R. Ali, M. Afzal, M. Sadiq et al., “Knowledge-based reasoning
building information modelling: from systematic literature and recommendation framework for intelligent decision
review to a methodological approach,” Building and Envi- making,” Expert Systems, vol. 35, no. 2, Article ID e12242,
ronment, vol. 182, Article ID 107164, 2020. 2018 Apr.
[19] V. V. Tyagi, A. K. Pandey, D. Buddhi, and R. Kothari, [36] A. Anvari-Moghaddam, H. Monsef, and A. Rahimi-Kian,
“Thermal performance assessment of encapsulated PCM “Optimal smart home energy management considering en-
based thermal management system to reduce peak energy ergy saving and a comfortable lifestyle,” IEEE Transactions on
demand in buildings,” Energy and Buildings, vol. 117, no. apr, Smart Grid, vol. 6, no. 1, pp. 324–332, 2015.
pp. 44–52, 2016.
[20] A. S. O. Ogunjuyigbe, T. R. Ayodele, and O. E. Oladimeji,
“Management of loads in residential buildings installed with
PV system under intermittent solar irradiation using mixed
integer linear programming,” Energy and Buildings, vol. 130,
pp. 253–271, 2016.
[21] S. Sharma, Y. Xu, A. Verma, and B. K. Panigrahi, “Time-
coordinated multi-energy management of smart buildings
under uncertainties[J],” IEEE Transactions on Industrial In-
formatics, vol. 1, 2019.
[22] A. Singh, M. Syal, S. Korkmaz, and S. Grady, “Costs and
benefits of IEQ improvements in LEED office buildings,”
Journal of Infrastructure Systems, vol. 17, no. 2, pp. 86–94,
2011.
[23] K. Lan, S. S. Kelley, P. Nepal, and Y. Yao, “Dynamic life cycle
carbon and energy analysis for cross-laminated timber in the
Southeastern United States,” Environmental Research Letters,
vol. 15, no. 12, Article ID 124036, 2020.
[24] A. Ndiaye, F. Locment, A. D. Bernardinis, M. Sechilariu, and
E. Redondo-Iglesias, “A techno-economic analysis of energy
storage components of microgrids for improving energy
management strategies[J],” Energies, vol. 15, 2022.
[25] D. Odd, “Know your buildings to make huge energy savings
[J],” Waste Management & Environment, vol. 25, no. 11, 2015.