Received 25 November 2022, accepted 13 December 2022, date of publication 21 December 2022,

date of current version 5 January 2023.

Digital Object Identifier 10.1109/ACCESS.2022.3231444

Artificial Intelligence Enabled Demand

Response: Prospects and Challenges
in Smart Grid Environment
MUHAMMAD ADNAN KHAN 1 , AHMED MOHAMMED SALEH 2, MUHAMMAD WASEEM 1,3 ,

AND INTISAR ALI SAJJAD 1 , (Member, IEEE)

1 Department of Electrical Engineering, University of Engineering and Technology Taxila, Taxila 47050, Pakistan
2 Electrical Engineering Department, University of Aden, Aden, Yemen
3 School of Electrical Engineering, Zhejiang University, Hangzhou 310027, China
Corresponding author: Ahmed Mohammed Saleh ([email protected])

ABSTRACT Demand Response (DR) has gained popularity in recent years as a practical strategy to increase
the sustainability of energy systems while reducing associated costs. Despite this, Artificial Intelligence (AI)
and Machine Learning (ML), have recently developed as critical technologies for demand-side management
and response due to the high complexity of tasks associated with DR, as well as huge amount of data
management to take decisions very near to real time implications. Selecting the best group of users to
respond, learning their attitude toward consumptions and their priorities, price optimization, monitoring and
control of devices, learning to engage more and more consumers in the DR schemes, and learning how to
remunerate them fairly and economically are all problems that can be tackled with the help of AI techniques.
This study presents an overview of AI approaches used for DR applications. Both the Artificial Intelligence
and Machine Learning algorithm(s) are employed while discussing commercial efforts (from both new and
existing businesses) and large-scale innovation projects that have applied AI technologies for energy DR.
Different kind of DR programs implemented in different countries are also discussed. Moreover, it also
discusses the application of blockchain for DR schemes in smart grid paradigm. Discussion of the strengths
and weaknesses of the evaluated AI methods for various DR tasks, as well as suggestions for further study,
round out the work.

INDEX TERMS Artificial intelligence, blockchain, demand response, demand side management, demand
response, Internet of Things (IoT), smart grids, machine learning.

I. INTRODUCTION systems, they possess greater challenges to operation and

Power grid operators face significant new hurdles due to management [1]. Integrating fast-response electrical power
the emerging growth of renewable energy sources (RES). production system, demand-side management, and other
There must be a quick adjustment in energy infrastructure energy storage devices are the primary methods for offering
to meet the growing diversity of integrated energy sources. flexibility [2]. Also, the digitalization converts conventional
RES in general are notoriously hard to estimate when it grids to modern smart grids to manage electrical grids. Power
comes to their power production because of their inherent systems may be made more efficient, secure, dependable,
instability and intermittent nature (i.e. they are totally robust, and sustainable with the use of new technologies
dependent upon variable solar and wind energy). Because like the IoT, real-time management and control of power
of the requirement for flexible power system to ensure at different level and smart contracts [3]. Several nations
the continued and safe functioning and stability of power have set ambitious goals for the widespread implementation
of advanced metering infrastructure (AMI) [4], [5]; for
The associate editor coordinating the review of this manuscript and instance, the UK’s Ofgem has mandated the installation of
approving it for publication was Sangsoon Lim . 53 million smart meters for electricity monitoring and control

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 11, 2023 1477
 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

2020 [6]. There has to be automated methods for analyzing

the enormous amounts of data produced by this infrastructure
(IoT, AMI). The evolution into faster, decentralized, and
intricate power systems [7] also brings forth new challenges
that may eventually become unmanageable for manual
control system. Artificial intelligence techniques have been
suggested as an important method for dealing with these
issues in power grids. More stable and efficient power system
may be achieved via the application of AI to forecast power
demand and production, optimize the maintenance and use
of energy assets, get a deeper understanding of energy
consumption patterns, and more. AI play its crucial role
in human life because it makes decision making process
easier and very near to the actual/real-time decision also FIGURE 1. Number of articles published related to different AI techniques
used in DR.
automatically schedules different appliances being used for
different purposes at domestic and industrial level.
and directions for future study. In addition, we believe our
A. MOTIVATION AND SCOPE OF THE REVIEW evaluation will be a valuable resource for future scholars
Although AI methods have long been studied and utilized and professionals in the subject. Specifically, this entails
in many power system applications in different areas, but educating them about the AI strategies that have proven
currently researchers paid attention about the application effective in solving problems similar to theirs in terms of
of AI in the context of demand response. Increasing the DR. Specifically, this involves a methodical evaluation of the
scale and scope of DR initiatives is an important factor for pros and cons of using a certain AI method in each particular
power system operators as it has been highlighted as one field of use. Finally, we wanted to provide more context
of the potential techniques to allow more demand flexibility than is provided by just reviewing academic publications by
to the power system. For DR schemes to perform better, discussing some of the research efforts in this field that are
an automated framework that can be more flexible and supported by industry. Our investigation reveals a total of
learn about its context (such as customer preferences) is forty businesses/commercial efforts and twenty-one major
necessary. In reality, it is becoming clearer that AI may projects in this space, demonstrating the high level of activity
contribute significantly to the future success of DR schemes and investment in this area. To the authors ’ knowledge, this
by automating the process while learning the preferences is the most extensive analysis to date of the use of artificial
of end-use customers. This framework is only possible with intelligence (AI) in the field of energy demand side response.
inculcating AI techniques for demand side management.
The dramatic uptick in study within this field is indicative B. RELATED REVIEWS ON DEMAND RESPONSE (DR)
of the growing need for AI-based solutions within the DR The literature on energy-demand-response reviews is abun-
industry. As can be observed in Figure 1, between 2011 and dant. In [8] looked at the Several advantages of DR
2021, the number of scholarly articles on this topic increased in smart grids which consists of different smart sensors
dramatically. As a result, there is a growing need for a used for control, monitoring, and communication systems.
comprehensive evaluation of the many artificial intelligence Innovations in DR systems, load forecasting approaches, and
(AI) algorithms now in use throughout DR’s many different communication channels are studied in [9]. Long-term, less
application domains. While many of these papers make obvious effects of DR, such as its influence on energy market
important contributions, they typically only examine one pricing and on customers, were investigated in [10]. In [11]
AI/ML method and one application area at a time. Based looked at the technical side of DR for frequency regulation,
on the body of information presented so far in current while economic effect of DR are investigated in [12].
publications, we believe there is a pressing need for an In addition, the optimum management of DR techniques and
in-depth analysis that charts the growth of the field and serves DR pricing schemes were improved in [13].
as a guide to the most promising AI approaches utilized in Particularly in [14] there is a literature analysis on different
certain sub-areas of DR. Because of this, the goal of this study Artificial Intelligence (AI) techniques for DR, discussing
is to give a comprehensive analysis of the different AI data- how AI might be used to create a scheduled monitoring
driven methodologies for DR applications. Our evaluation system for a Smart home that is being monitored under
has three main purposes: DR. For example, [15] compared and evaluated several self-
At first, we want to provide a thorough review of the organizing optimization algorithms for demand response
artificial intelligence (AI) methods underlying this field, in smart buildings while in [16] author used clustering
as well as the primary particular applications/tasks in power technique to mitigate the load profile and managing load
DR in which these methods have been applied. Therefore, demand intelligently and [17] narrowed their attention to
providing a holistic view of the development of the area the use of reinforcement learning for DR specifically.

1478 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

In addition, [18], explored the use of smart meter and its data of AI techniques in demand side management specially
analysis for DR application while [19], [20] examined AI for DR like for forecasting techniques, Load management
based load prediction, concentrating mostly on deep learning or Energy management. DR programs being implemented
and artificial neural networks (ANNs) [21]. aggregation of in different countries are explained in section 5 while
thermal inertia, particularly from district heating networks is blockchain application on DR in smart grid is explained in
emphasized in [22] and [23] highlights the emerging concept detail in section 6. While challenges and future scope is given
of integrated demand response, which integrates multiple in section 7 which followed the conclusion in section 8.
energy types and vectors (including electricity, natural gas,
and heat). Our main focus, on the other hand, is on electricity II. DR RESOURCES, OPEARTION, AND MARKET
demand and demand side management that goes into further STRUCTURE
detail on the artificial intelligence methods that make this The conventional design of the power grid is based on a
possible. one-way flow of energy from generation end to consumer
While the above-mentioned assessments of AI technolo- end. High voltage generators under centralized control are
gies for DR applications have proven helpful, it is worth employed to provide this supply. Demand side management
noting that they are often limited in scope. They often and notably demand response have emerged as viable options
focus on one particular facet of artificial intelligence, such for the efficient and reliable operation of the electric grid as
reinforcement learning [17] or one area of application, like a result of the proliferation of grid service markets and the
home energy management systems [14]. The goal of this inclusion of DER in recent years. However, a DR model,
research is to present a more all-encompassing and global in contrast to conventional power grids, calls for a two-way
perspective on the AI methods now used in DR schemes communication system and intelligent algorithms to analyze
to facilitate power system function. To identify possible the produced data. Because of this, smart meters are a
research gaps and propose future study directions in this crucial part of a smart grid and play a pivotal role in DR
rapidly expanding domain, we believe that a systematic models [24]. Information gathered may also be used by
review of this size and breadth is necessary and desirable. AI-based solutions to improve DR initiatives. The purpose
of this section is to provide an overview of DR services and
define their place in the existing framework of the electrical
market.
Demand side management is a broader area in electrical
power system of which demand response is a subset, and is
achievable because of emerging technologies and innovations
in conventional grids i.e., due to smart grid operations [25].
When we talk about ‘‘Demand Response’’ in this article,
we’re talking to the ways in which commercial and industrial
consumers of electricity alter their consumption patterns.
Customers agree to modify their typical load profile as
required from the utility provider in order to get maximum
possible benefits for both sides by decreasing or shifting
their energy usage away from peak hours [25]. Although
DR encompasses a wider range of energy sources (such
as thermal energy, gas, etc.), this study will just discuss
electrical power aspects of the generation side. Figure. 3.
Shows the classification of DR, where DR programs based on
motivation offered to customers are further divided into two
distinct categories, which corresponds with the classifications
provided by [9], [26], and [27].
FIGURE 2. Layout of the paper.
A. PRICE BASED DR PROGRAMS
With the goal of getting end-use consumers to shift their
C. STRUCTURE OF THE REVIEW energy consumption habits, the electricity price in this
This paper’s structure is given in Figure 2 and is organized scenario fluctuates over different time intervals. Time-of-use
as follows. Section 2 will first introduce DR and its pricing, critical-peak pricing, and real-time pricing are all
connection to the electrical grid and energy markets. The examples of such schemes [10].
next Section3 introduce AI and its different techniques
like machine learning, deep learning and other techniques B. INCENTIVE BASED DR PROGRAMS
under AI and its fundamental ideas before classifying the End-use customers are incentivized to lower their power
evaluated literature. Section 4 describes the applications use via these programs by providing them with offers to

VOLUME 11, 2023 1479

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

FIGURE 3. Classification of DR resources, operation and structure.

do so upon request or in accordance with a contractual company, depending on the national policy. The items may be
agreement. Direct-load controls (DLCs), interruptible tariffs, exchanged on the spot market (day ahead and/or intra-day) in
and demand-bidding programs are all examples of this kind both circumstances and for ancillary service in spot market.
of scheme [28]. If a resource provider makes a definite promise to provide
Each of these methods of management calls for tailoring a particular quantity of electricity into the grid, they must
the incentives or contracts offered to customers in light fulfil that promise or face financial consequences. Therefore,
of their actual behavior. There are also two distinct types it is crucial for DR monitoring companies to agree the end
of markets for electricity: retail, where electricity retailers consumer and offer power flexibility.
directly contract with end-users for electricity delivery, As distributed energy resources (DER) become more
and wholesale, where retailers, suppliers, producers, grid integrated into power networks, innovative approaches are
operators, and third-party aggregators all work together to needed to meet the technological constraints of a smart grid
ensure that retailers can deliver power to their customers (mainly frequency and local voltage regulation). DR is one
without jeopardizing the reliability of the grid. Three distinct of the best options. Smart DR plans may be built upon the
markets—the energy market, the capacity market, and the infrastructure that is being put in place with the usage of
ancillary services market—make up the wholesale electricity smart meters for homes and commercial buildings along with
market and work together to incentivize various parties to IoT-enabled smart appliances being deployed in smart homes.
contribute to the nation’s power distribution and the grid’s Furthermore, in order to effectively regulate demand without
efficient operation and reliability. The energy and auxiliary adversely impacting end-user comfort, these tactics will use
service sectors are linked with demand-side responsiveness. AI-based smart algorithms.
Contracts between market participants may be conducted In the next part (Section 3), we describe different artificial
bilaterally (over the counter (OTC)) or by a proper broker intelligence (AI) methods offered and researched in literature

1480 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

in the interest of automating DR. Section 4 then discusses the to the scarcity of labelled data, this is useful in DR. Clustering
applications of AI and ML to various DR services. has been the primary use of unsupervised algorithms in
DR, whereby items (such as load profiles) are grouped
III. AI APPROACHES/TECHNIQUES IN DEMAND together such that their members are similar to one another
RESPONSE but different from those in other clusters. Consumers have
AI, or artificial intelligence, is the study and development been categorized, and typical load profile shapes have been
of intelligent entities (agents) [29]. These smart agents are identified, with the help of several clustering techniques.
computer programs that can analyze their surroundings and Therefore, this categorization may be used to choose
take appropriate action to accomplish certain objectives. consumers for DR programs, pay consumers for participation
Computer science, neurology, economics, information the- in DR programs, and identify families who can benefit from
ory, statistics, psychology, control theory, and optimization DR schemes.
are just few of the many disciplines that contribute to AI. The concept of ‘‘interactional learning’’ is central to
Hence Agents with artificial intelligence (AI) may vary almost all learning philosophies. To learn from experience,
from robots with true thinking capabilities to search algo- Reinforcement Learning (RL) is one of the most fascinating
rithms utilized in board games. Different methods have computer methodologies. RL is a method that considers the
been used since the advent of AI in the 1950s to develop whole challenge of an agent learning to achieve a certain
intelligent computers. Among these methods are statistical objective in the context of an unknown environment [35].
learning [30], [31] soft computing [32] and knowledge- The two most distinguishing features of RL are the use of
based systems [33] In this work, we will examine the data- a trial-and-error search strategy and the provision of delayed
driven, soft-computing, non-symbolic approach to AI. This reward. RL has been used from a very long ago for DR to
study also examines AI methods in both the single-agent and monitor and control the various kinds of loads at domestic
multi-agent contexts in order to provide a more complete level or Electric vehicles while taking into consideration the
picture. Figure 4 shows many types of AI approaches that customer preferences. RL technique proved to be on of the
have been used to DR and how they have been classified. best and easiest model for DR to manage the complex data
for both customer and service provider. Researchers have also
A. MACHINE LEARNING AND STATISTICAL METHODS utilized the RL framework in order to estimate the budget for
In the age of big data and the Internet of Things, automated utility [36], [37], [38] and to create a scalable and efficient
analysis of the ‘‘data tsunami’’ that is variable constantly model for a group of customers [39].
being generated is crucial. Machine learning is an essential
part of artificial intelligence (AI) that consists of a collection B. NATURE-INSPIRED ALGORITHMICS
of techniques that aim to learn from data. Methods that auto- When developing new computational methods, scientists
matically recognize patterns in data and utilize these patterns have always looked to natural and biological systems for
to make predictions and other kinds of decision making in an inspiration. Artificial intelligence (AI) researchers have used
uncertain environment fall under the umbrella of this set of algorithms inspired by nature to perform tasks including
AI approaches [34]. Machine learning is an interdisciplinary looking for relevant information and figuring out the best
field that largely uses ideas from computer science, statistics, course of action to achieve a goal [29]. Meta-heuristics
mathematics, and engineering. Murphy [34] lists supervised inspired by evolution, biological swarming, and physical
learning, unsupervised learning, and reinforcement learning processes are the most common types of algorithms drawn
as the three primary categories of machine learning. from nature. The term ‘‘meta-heuristics’’ is used to describe
With an already labelled collection of input-output pair- a group of iterative processes that supplement heuristic
ings, the purpose of supervised learning is to learn a procedures through the application of proper hieratical flow
mapping between the input vector x and the outputs y. The model to find the most optimum and efficient solution
inputs xi might range from a basic real number to a highly using different kind of nature inspired intelligent learning
structured object, and this collection of data is known as strategies [40].
the training set (e.g. an image, a timeseries, a graph, etc.). Algorithms inspired by nature have been utilized exten-
By using kernel-based and tree-based approaches in addition sively in the field of DR, mostly for consumer-level load
to linear regression models, supervised learning techniques scheduling (algorithm included in HEMS) and for assisting
have been largely employed in DR to anticipate the demand aggregators and retailers in optimizing the price of their
and power pricing. Supervised ANNs are also widely used for DR service provider clients. In the DR setting, where the
forecasting purposes. scheduling problem might be computationally costly, meta-
Using unsupervised learning techniques, the system is heuristics have been widely used because they can discover
merely provided with the inputs and is tasked with finding solutions in a fair amount of time.
potentially relevant patterns of interest within them. Because Evolutionary algorithms, also known as Evolutionary
the patterns that need to be recognized are unknown in Computation (EC), are a heuristic-based technique that
advance and there are no clear error measures to employ, mimics key aspects of biological evolution in a computing
unsupervised learning is a less well-defined technique. Due setting, including different kind of biological processes

VOLUME 11, 2023 1481

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

FIGURE 4. Different artificial intelligence techniques used for DR.

like reproduction, mutation, recombination, and selection. examples of swarm intelligence algorithms in the research
The process is evaluated for different number of iterations literature. For further details on these algorithms, see the
until the optimization function on an individual reaches a reviews [51], [52], [53]. Swarm AI systems may become
termination criterion. Genetic algorithms (GA) are biological trapped in local optima and have a sluggish convergence
model inspired by natural processes in humans, animals pace, much as evolutionary approaches [40]. In contrast to
and birds according to Charles Darwin’s theory of natural GA, where ‘‘poor’’ particles are eliminated, all particles’
selection [41], [42], [43], [44], [45] and have emerged as histories are used in swarm AI systems to aid in the
the dominant approach from the evolutionary computation search [54]. In addition, swarm AI approaches often have
in the literature on energy DR [43], [44], [45], [46]. fewer parameters that need fine-tuning before deployment.
Implementing these meta-heuristic genetic algorithms gives Swarm AI algorithms are often employed by energy aggre-
maximum possible efficiency for turning different appliances gators and retailers to determine the best scheduling and
ON/OFF at domestic and industrial level. For example, pricing strategies to minimize costs associated with demand
a differential Evolutionary Algorithm (EA) which is utilized response. The optimization issues in DR are notoriously
in battery management system in order to manage the data non-convex because they include a high number of variables,
of lithium-ion battery in datacenter [47], and a bi-level quadratic optimization functions, and limitations derived
Evolutionary Algorithm (EA) which is used to find the from the calculation of the AC power flow. Heuristic
maximum efficient electricity tariff for consumers under DR optimization is well-suited to this situation since it can
strategies [48] are all examples of evolutionary algorithms quickly locate a near-optimal solution while requiring less
being used in the DR setting. effort than other mathematical methods. A most popular
Swarm intelligence is a branch of artificial intelligence example of these heuristic optimization methods in DR is
that studies how imitating the behavior of biological swarms particle swarm optimization (PSO).
might help with problem solving [48]. The Particle Swarm In addition to these algorithms, several nature-inspired
Optimization (PSO) method [49] and the Ant Colony meta-heuristics have been discovered, many of which defy
Optimization (ACO) algorithm [50] are the two most popular easy categorization. With inspiration from the mechanisms

1482 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

present in biological immune systems [55], the CLONALG- D. MULTI-AGENT SYSTEMS

based [56] Artificial Immune System (AIS) algorithm is Because the demand side of power systems is often decen-
utilized to set prices for the aggregators. The Wind Driven tralized, there is a pressing need for methods that can learn,
Optimization (WDO) algorithm [57], which is based on plan, and make choices in the context of a system comprised
atmospheric motion, is used to find an optimum schedule of of various interacting intelligent agents. Multi-agent systems
appliances at the household level, and the simulated annealing (MAS), a branch of distributed artificial
approach, which was developed on the annealing concept, is intelligence, give the analytical tools necessary to investi-
used for DR. gate such issues. This review focuses on the subfields of MAS
known as automated negotiations, which focus on finding
C. ARTIFICIAL NEURAL NETWORKS a common agreement between different participants in the
Computer models called Artificial Neural Networks (ANNs) game/scheme, cooperative/coalitional game theory, which
take cues from the structure and function of real-world focuses on the study of coalitions among these players and
neural networks. This study presents ANNs as a separate designed agreements.
category because to their widespread use in DR applications; One important idea of Game Theory is ‘‘game,’’ which
nevertheless, ANNs might technically be classified as either is represented by a mathematical model that ‘‘captures the
a machine learning method or a method inspired by nature essential aspects of the interaction between self-interest
in the field of artificial intelligence. Multiple fields have entities’’ [66]. Understanding what makes a result of a game
made use of ANNs for tasks such as classification, clustering, logical is a central goal of game theory, and many solution
pattern recognition, and forecasting [58]. In DR, ANNs ideas have been devised to help narrow down the possibilities.
have used for load forecasting, and they have been of One such notion is the Nash Equilibrium.
varying designs and complexity (number of layers). ANNs A subfield of game theory known as ‘‘coalitional’’ or
are used in the majority of DR applications to predict the ‘‘cooperative’’ game theory in which the aim is to foresee
future consumption of an infrastructure (building, appliance, which coalitions will emerge. As opposed to focusing on the
consumer group), the flexibility of a load, or the short- strategies used by each player individually, cooperative game
term pricing of power (which can be for hours, minutes theory [66] instead divides the payout among the participants.
and even world is progressing towards seconds to one day Cooperative game theory has seen extensive use in the DR
ahead). In certain cases, ANNs may serve as a viable setting, particularly in instances where legally enforceable
alternative to nonlinear regression methods. For instance, agreements have been established (i.e. incentive-based DR).
most load forecasting implementations takes inputs like Selecting the best group of power users to take part in
previous consumptions, weather, day, hour and sometimes the DR schemes and dividing up the coalition’s payout are two
price. Inputs for a price forecast are often prices from the primary uses of cooperative game theory in DR (known as
past. A lot of literature has led to the identification of two solution concept). Depending on the requirements that the
primary classifications, single hidden layer ANN and Deep aggregator is trying to satisfy, the solution idea relates to
Learning. the method in which the income is divided in between the
Single hidden layer, feedforward ANN has wide range participants in DR for power demand flexibility.
of application in the DR sector. For the most part, single Strategically speaking, Mechanism design is a subset of
hidden layer ANNs have been used in the DR literature social choice theory which postulates that different kind of
for load and pricing forecasting. In addition to these participants would act in a manner that maximizes their
applications, single hidden layer ANNs have been utilized to own utility. Since the ability to ensure specific qualities
represent complicated functions like cooling system control is crucial to the maximum possible optimum solution
algorithms [59] or urging users to manage and shift the load for DR schemes, mechanism design has been frequently
to the off-peak hours [60], But this can be directly linked to employed in DR literature. Consumers are encouraged to
the day temperature, weather and electricity cost. submit accurate bids using incentive-based mechanisms
Deep learning is a kind of ML that processes data in designed with the help of mechanism design in DR.
its raw format and automatically discovers the data which Several articles [67], [68], [69] suggest DR techniques that
is needed to be represented for detection or classifica- guarantee consumers will optimize their utility function by
tion [61]. It involves learning several layers of representation providing accurate reports of their preferences. Incentive
and abstraction. Like single hidden layer ANNs, deep compatible mechanisms (IC) are those that may be used
architectures have mostly been used for load and pricing in conjunction with incentives. Future pricing and end-user
forecast in DR. In addition, deep architectures have been choices for various time periods throughout the day inform
used for customer response prediction [62], DR event- the scheduling and payment function proposed in [68].
aware home appliance control [63], identifying consumer Two ‘‘penalty-bidding’’ mechanisms based on a dominant-
socio-demographics to inform targeted DR mitigation strate- strategy equilibrium are presented in [69], whereas [68]
gies [64], and customer clustering based on the estimated offer a mechanism that takes the opposite, ‘‘reward-bidding,’’
load curve which is already encoded in the system using deep approach. Last but not least, [70] present a cooperative
learning based autoencoders [65]. mechanism that is both efficient and incentive compatible,
VOLUME 11, 2023 1483
 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

meaning that participants do not benefit from increasing 1) LOAD FORECASTING

their baseline consumption in order to demonstrate a false Load forecasting and estimate are crucial to the safe and
demand decrease. The aggregator chooses a selection of effective functioning of any electricity grid. Proper demand
agents, submits bids on their behalf to the electrical flexibility forecasting is a crucial tool for addressing several DR
market, and then splits the proceeds with them based on their concerns, such as ensuring adequate planning, compensating
pledge to cutting down on energy use, while punishing those DR participants, and determining the capacity potential of DR
who actually raised their usage. resources [19]. Long-term load forecasting (>24 hours) and
Among a group of agents, products as in [71], resources as short-term load forecasting ( 24 hours) are commonly used
in [72], and tasks as in [73] may be divided up via negotiation. to categorize demand forecasting. The publications included
The existing research classifies allocation processes into two in this overview are those that focus specifically on the load
broad categories ‘‘Auctions’’ and ‘‘Negotiations’’ [74]: forecasting issue in the DR context. The analysis in [19]
• Auctions are systems wherein a group of people compete provides a more comprehensive look of load forecasting
with each other in an automated process controlled by a third in the smart grid setting for anyone interested in learning
party. Here, the norms and the procedure are set in stone. more.
Using concepts of mechanism design, the goal of auction There are a number of publications in the literature that
theory is to arrive at an optimum auction model which ensures attempt to predict demand, some of which account for poten-
a set of desired qualities. tial changes in demand brought about by DR. Most of the
• The goal of any negotiation is to reach an agreement studies [77], [78], [79], [80], [81], [82], [83], [84], [85], [86]
between two or more parties through an exchange of informa- focus on estimating demand for the next day or two, while
tion that includes offers, counteroffers, and arguments [239]. others [87] look forward a week. Also, load forecasting has
Negotiations encompass a diverse and an unclear set of been done at many other aggregation levels, including for
iterations which are used to distribute goods, resources, single-family homes [78], [88], [89], [90], commercial build-
services, or tasks. More complicated and individualized ings [80], [83], [91] and individual appliances [92], [93] such
agreements, as well as more decentralized and adaptable as chillers, ice banks, and lights. The load forecasting for sin-
protocols, are all possible with the help of automated gle consumer or a group of consumers for day ahead predic-
negotiating methods. tion depends on previous load profile and weather conditions
as presented in [92] using ANN-based method for home load
IV. APPLICATION AREAS OF AI IN DEMAND RESPONSE
forecasting. Artificial neural networks may also be used for
Many electrical power system factors, including load and appliance-level load predictions. Deep neural networks with
energy price forecasts, selecting the most appropriate cus- a principal component analysis (PCA) based feature selection
tomers for DR schemes, and developing automated systems strategy to estimate loads of home appliances is presented
for managing demand-side resources, must be considered for in [93], whereas [83] use ANNs to anticipate loads of HVAC
the successful implementation of DR programs. Forecasting, systems.
real-time management of networked infrastructure, taking Baseline load estimate describes the scenario when load
best decision very near to optimal decision, adapting to a forecasting is performed without considering DR. In the
dynamic behavior, learning from load profile and many other context of DR, the baseline load can be determined as the load
areas where AI technologies have been used in DR [75]. Here, that is feeding power in the absence of the DR programs. [94].
we categorize the different applications of AI in the field of Rewarding DR participants requires accurate estimates of
DR that have been found in the published literature. typical power usage, which may be obtained from the base-
A. FORECASTING IN DR line consumption measurement of consumers [94]. Baseline
Forecasting has been one of the primary applications of AI load estimate research for homes [94], [95], commercial
methods. It has been discovered that artificial intelligence buildings [96], [97], and industrial plants [98] may be found
(AI) techniques are being implemented for forecasting in the aforementioned works.
of power prices for different kind of loads in the DR In addition to the above, the practice of flexibility
environment. Schedules for delivering power in the near forecasting has been the subject of several research inves-
future may be influenced by forecasts, and longer-term tigations. A flexible load is one that can be adjusted in
plans for the system and service providers can also benefit response to changes in the time of day, the weather, and the
from forecasts [76]. Improving power scheduling with smart-grid control signal [99]. Studies have estimated the
short-term forecast helps aggregators provide better services DR heat-load flexibility of homogeneous [100], [101], [102]
and customers react more closely to appropriate DR signals. and heterogeneous [103] VPP heat-load clusters. Both the
If service providers and operators have more accurate long- DA market and other energy markets are open to trading
term estimates, they will be better equipped to make decisions the estimated flexibility [100], [103]. Research in this area
about how much flexibility to provide, which customers to includes energy flexibility prediction, forecasting of smart
prioritize for DR, and how much to compensate or charge homes air conditioning systems [98] and peak time DR
those customers. capacity calculation [104].

1484 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

2) PRICE FORECASTING primary techniques used in prediction techniques and noting

Electricity price forecasting has been done at both the their advantages and disadvantages. Wavelet transform is
aggregator and the end-user levels. In a multi-aggregator employed to convert the initial data into its small component
context, in which out of group of customers, only one to smooth out the peaks and valleys in the initial data [124].
consumer is implementing a DR scheme is described in [105] Several research [121], [122], [123] have used this technique.
for predicting the regional wholesale energy price from the The DSO and BRPs have a difficult time coping with
demand bids of the different aggregators to the SO. The ideal distributed solar energy due to a lack of information on the
incentive rates for various customers are determined in part prosumers’ aggregated small-scale solar power. To get around
by a model employed by Lu and Hong [106] to anticipate this issue, [124] makes guesses about the total power output
the price of electricity on the wholesale market. The bulk of of unmonitored distributed solar installations on residential
the publications [107], [108], [109] focus on the consumer roofs.
level and attempt to predict the residential load a day before
in order to manage the production under the influence of price 4) FLEXIBILITY FORECASTING
incentives. In contrast [110] predict the hourly fluctuations in Flexible loads, decentralized storage facilities, and (DERs)
power costs for businesses. may all be combined into one large pool of potential
flexibility, and the advent of aggregated flexibility prediction
3) GENERATION FORECASTING makes it possible to define the default area’s share of this
The purpose of the generation forecasting is to anticipate resource. Aggregated distribution grid flexibility may be
the amount of power that will be generated or required used to lessen the demand for grid expansion and improve
to be generated from renewable sources. Better Renewable the efficiency of power system by improving power system
Energy Systems (RES) projections has grown with the operations [125]. The computation of aggregate flexibility is
inclusion of demand response by implementing proper DSM elaborated on in [126].
techniques to offer more flexibility to power system with In order to (i) give flexibility facility to power system
the rising renewable energy resources. In [111], the authors agents (ii) reduce end-user energy bills through Home Energy
discuss current research that assesses the efficacy of artificial Management Systems, and (iii) participate in electricity
intelligence (AI) approaches used to RES prediction models, markets by using optimal bidding strategies [127], [128] the
considering a wide variety of forecasting horizons, and a aggregator is the service provider responsible for gathering
wide variety of renewable energy generating sources. Results and controlling its portfolio of flexibility sources [128] and
show that benchmark ML models can process big datasets develops new flexible business model methods. Demand-
and provide reliable forecasts; however, by integrating side flexibility aggregation becomes crucial for balancing the
machine learning algorithm models can enhance the system future power system due to the growing penetration of inter-
reliability. Two reviews cover different aspects of solar mittent RES and the sizeable number of residential users with
power forecasting: [112] discusses PV forecasting using potential flexible sources [128], [129]. In [125], ML-based
ML and metaheuristic techniques, while [113] zeroes in on regression models are used to forecast residential cus-
time-series statistical, physical, and ensemble approaches. tomers’ flexibility for real-time applications. [130] presents
In [114], the authors examine the current status of SVM as a framework for flexibility forecasting and how it can be
it pertains to solar and wind forecasting. Although the SVM managed across various energy sources and domains. [131]
regressor is straightforward and accurate, it does not scale decides to construct a flexibility load forecasting model
well to big data sets and performs poorly with high levels of for DR capacity scheduling using the GBM ensemble
noise. technique.
In [115], a variety of ML techniques are investigated for The aggregator provides a financial incentive signal
use in the production of wind and solar electricity. In [116], in [132] with the intention of encouraging a shift in demand-
the authors analyze several PV forecasting methods, from the side consumption. [133] estimates the adaptability potential
most basic to the most complex, and draw the conclusion of wet appliances (dishwasher and washer machine) in
that certain methods work better than others in different France. Residential load flexibility projections are com-
climates. Models for PV forecasting one day in advance puted in [133] using the NILM method. [134] considers
using deep learning neural networks are the subject of [117]. predetermined customer preferences, loads, and PV forecast
In [118], we take a look at how CNN may be used for uncertainty to define a workable flexibility space from
multi-site PV forecasting. In [119], the most prominent ML controllable home resources. The proportion of end users who
techniques for predicting wind speed and power are reviewed. must be able to submeter in order to determine the aggregate
These techniques include data preparation, initialization of demand composition is examined in [104]. According to the
different parameters, their optimization, and at the final stage findings, only 5% of submetering coverage is necessary to
of error minimization. These hybrid methods often provide accurately predict the aggregated load composition at the
better results than using individual models alone. In [120], substation level. [135] proposes a scalable and non-intrusive
the authors examine the use of Artificial neural network methodology for determining the flexibility of thermal
for variable wind energy production, bringing together the loads.

VOLUME 11, 2023 1485

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

B. SCHEDULING AND CONTROL OF LOADS FOR DR for residential appliances that accounts for limitations across
For service providers and their customers, the sheer variety appliance classes.
and quantity of DR devices available presents a significant Minimizing electricity cost [45], [57], [125], [148], [149],
issue. It is not practical for service providers to manually [150], [151], [152], energy consumption [57], [63], Peak
schedule and control their DR units throughout their portfo- to average ratio (PAR) [57], [150], and maximizing
lio. Further, automating the scheduling and management of social welfare [153] and reducing environmental pollu-
the many demand-side appliances is crucial for increasing tion [154] are common goals in the monitoring and control
consumer engagement in DR schemes, since otherwise aspects of customers’ loads. These goals must be achieved
customers would experience response fatigue [136] and stop while taking into account the tastes of the consumers.
participating in the DR program altogether. When it comes to There are two primary ways to determine what people
DR, scheduling and management of the different units may want. An intelligent algorithm, which may be either pre-
be handled either by the service provider (aggregator) or by defined [144], [145] or learnt [155], [156], can be used to
the individual user (consumer). The size and breadth of the reflect human preferences for the operation of household
units used distinguishes between the two tiers as the primary appliances. The second strategy involves putting limits on
distinction between them. When it comes to scheduling and what is considered a realistic timeline [149], [150], [151].
controlling devices at the aggregator level, the algorithms TCLs, such as heat pumps [144], [148], [157] and water
utilized must be more flexible and adaptable to a wider variety heaters [144], [157], air conditioners [144], [158], [159],
of circumstances than those used at the consumer level. battery storage systems [144], [160] and electric vehi-
cles [144], [160], [161], [162] are common examples of
appliances that are controlled in a DR setting. Although
1) LOAD SCHEDULING AND CONTROL AT THE
most papers [150], [151], [155] focus on residential buildings
AGGREGATOR LEVEL
as consumers, there is a lot of research literature focusing
While DR unit control is intuitive, improving solution
on scheduling like for scheduling of different loads at
time efficiency requires careful consideration of how and
small commercial buildings level is presented in [162],
when certain events will occur in a scheduling issue.
while charging points for smart Electric Vehicles may got
The scheduling process may be seen as a multi-objective
overloaded at peak hours, so they need to schedule their loads
optimization problem with constraints. A huge amount of
and customers while considering the customer level load
literature data is available on scheduling DR resources while
scheduling in order to provide a better DR service [162], and
considering no limitations [137], also including imposed
also at industrial level optimization model which consisted
network constraints [138] while keeping balanced system
of multivariable price function has been used to mitigate the
conditions [139] like [140] can be taken an example of
load problems [163].
planning the load a day before with proper monitoring
of generation sources. Additionally, [141] perform proper C. DESIGN OF INCENTIVE/PRICING SCHEME FOR DR
management for the forecasted DR units for DA in order
Both the aggregator’s and retailer’s bottom lines and the DR
to increase utility provider benefit cost and users flexibility
scheme’s effectiveness are impacted by the pricing or incen-
while minimizing fluctuation from variable resources (like
tive mechanism’s design. A DR program’s ability to recruit
wind and solar) on the grid. For utilization and monitoring
new members and keep current ones engaged depends in part
of commercial and industrial loads flexibility [142] created a
on the quality of its reward system. When it comes to pricing
cooperative and decentralized agent-based platform that takes
mechanisms, the vast majority of papers utilize AI methods to
into consideration the dynamics of each building separately.
determine the best interactive scheme a day in a meritocratic
Scheduling the charging of EV fleets is also the subject of
electricity market [66], [80], [107], [108], [138], [139], all
study for the purpose of DR service provision [143].
while maximizing the profit of the service provider within
the bounds of realistic market constraints and the discomfort
2) LOAD SCHEDULING AND CONTROL AT THE CUSTOMER of consumers for load reduction/shifting. Further, [164] use
LEVEL the numerical model relationship between real-time total cost
Energy management systems (EMS) [57], [144], [145], are and energy expenditure in a DR scenario for real-time pricing,
composed of several systems that work together to automate and [110], [165] have constructed a model based on the
the scheduling of the different consumers and their loads at cost elasticity matrix which is proposed for dynamic pricing.
residential and non-residential buildings. Automatic choices A novel tariff structure, dubbed prediction-of-use (POU),
are made by EMS in response to DR signals, with consid- is proposed by [25], which determines the rate by comparing
eration given to power costs, customers’ lifestyle trade-offs, the end-use users expected and actual power consumption.
and the most efficient usage of appliances and equipment. Electricity prices, including upper and lower limits, are being
The implementation of DR systems by households and small investigated concurrently by [134] for a bi-level model of
businesses and factories depends on automated EMSs. Load consumer pricing.
scheduling for DR under an EMS is investigated in [146] Several articles [24], [26], [27] discuss the topic of how
and [147], proposing a user-independent power scheduler to appropriately compensate a group of consumers who

1486 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

are collectively lowering or shifting demand during a load

attenuation event via the use of incentive mechanisms.
In addition, Lu and Hong [109] analyze the profitability of
both users and service providers (aggregators) in a structured
energy market to determine the appropriate incentive rates
for various electrical users. A two-part reward function
is developed by [135] to incentivize DR participation,
with participants receiving rewards for their individual and
collective efforts to achieve a reduction objective. One portion
is the agent’s compensation for its role in the reduction
effort, and the other part represents any fees or fines it must
pay. Money is offered (a reward) to customers in return for
cutting down on consumption in a model developed by [115],
which simultaneously learns the likelihood that consumers
will take up the offer. To begin building a multi-round bidding
model, [141] learn the interruption load compensation price.
In [134] author also suggest a novel DR mechanism that uses
Vickrey-Clarke-Groves pricing to provide a customizable
suite of DR contract options. This novel system selects a
selection of consumers to cut down on consumption while
also factoring in the likelihood that the goal will be attained FIGURE 5. Role of deep learning techniques for DR programs in smart
grid.
(reliability). By including variable preparation costs, different
effort levels, and multi-unit consumption reduction, [132]
generalize prior work [166]. This paper suggests a reward- compensation rates. The most effective resources are well-
bidding technique as an alternative to a penalty-bidding compensated in this way, which encourages them to take part
mechanism for achieving effective incentives. Other research in the DR programs which provide them maximum possible
focuses on the design of contracts for incentive-based DR. benefits. The creation of DR programs and demand control
For example, [167] examine bilateral contracts (between strategies [168], the gathering of DR resources [177], [178],
a retailing agent and a business client) in a multi-issue the evaluation of a DR project’s potential benefits, and
negotiation context. In a similar way, [58] devise incentive the identification of hourly loads for carrying out DR
agreements for ancillary services, in which service providers programs [179] are other uses of classifying consumers. In a
engage in the commodity supplementary service market number of works, customers are categorized based on their
and coordinate engagement with end users at the retail quotation data [188], behavior [180] for EVs engaging in
level. Other than the two-way dialogue between the service DR, predicted effects of the DR program [181], number
provider and the customer, their work also considers the of residence occupants, building size, building type, and
interaction amongst customers. terrain type. While [182] partitioned the flexibility of EVs
for DR services by clustering EV charging sessions, in [183]
D. LOAD/CUSTOMER SEGMENTATION author designed flexibility envelopes of TCLs for DR. For
Segmenting/Grouping power users into categories is a DR, [184] combined the flexibility of batteries and modest
crucial application for DR which is an assistance to utility non - dispatchable loads using a unique clustering approach.
providers in creating DR programs, pooling services, assess- The overall flexibility of an aggregator’s portfolio of assets
ing the load capabilities of joining various DR programs, has been calculated using clustering approaches in the more
etc. [168]. The produced groups of consumers are created general situation of energy markets, such as the work in [185].
in the study literature to carry out various tasks in the
DR context. A significant portion of the evaluated works E. ENERGY THEFT DETECTION
classify customers to find potential participants for DR Theft of energy, or tampering with one’s electric power
programs [77], [169], [170], [171], [172] and identify the data, is done so that one may pay less for their electricity.
best group of consumers who are already enrolled in DR This is one of the biggest crimes ever committed in the
programs to contact in order to reduce demand during DR USA. Several methods exist to steal power in an SG.
programs [173], [174]. According to [64], load profiles As shown in Fig. 5, some of the methods include wiping
can be used to extract socio-demographic data, and the recorded events, altering stored demand, manipulating the
characteristics of these consumers can be used to choose meter, disconnecting the meter, and so on. In the past, public
potential DR participants. The typical daily load profiles reports would prompt power companies to dispatch teams
that are created for each group are then used in [175] to to inspect electrical infrastructure. Smart meters and other
create customized power price plans for price-based DR developments in the metering infrastructure have made it
programs. DR resources are grouped in [176] to determine simpler to identify instances of energy theft. This resulted

VOLUME 11, 2023 1487

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

in the development of AMI (Advanced Metering Infrastruc-

tures) [186]. AMIs, however, bring a number of drawbacks,
including the possibility of manipulated meter readings.
Because of this, a new feature engineering framework has
been developed, whose primary purpose is to prevent energy
theft from smart power networks. For market segmentation
purposes, there is a suggested framework in [187] that
uses both genetic programming and finite mixture model
clustering. The purpose of this was to produce a collection
of features that effectively communicates the relevance of
demand over time. In addition, the ability to compare results
from different houses gave it a high degree of accuracy
for spotting irregularities and fraud. Many distinct ML
algorithms were used. The amazing result may be attributed
to the computationally extremely practical nature of this
approach. The Gradient Boosting Machines outperformed
all previous ML classification models with respect to the
classification method. This may be explored more thoroughly
in future studies, and it has important practical implications
for power utilities.
If energy is distributed but never invoiced or paid for,
this is known as non-technical loss (NTL) in SGs. This
has become an international crisis in the electricity supply
sector. In [188] there is a suggestion of a concept based on a
power distribution network, the intermediate monitor meter, FIGURE 6. Application/areas of AI in DR.
to simultaneously detect bypassing of the meter and NTL of
meter manipulation (IMM). For a thorough analysis of the assess the credibility of each consumer’s claimed energy
power flow and efficient NTL detection, this model separates use pattern. The experimental results revealed that this
the network into finer and more autonomous networks. strategy significantly enhanced the pinpointing accuracy in
The energy balance between the IMMs and the collector comparison to other recently existing methods. The Internet
is analyzed, and a suggested method for NTL detection is of Things (IoT) and artificial intelligence (AI) are two crucial
created to solve the resulting linear system of equations enabling technologies for smart cities. For the purpose of
(LSE). The authors also described the IMMs’ underlying identifying energy theft, [191] suggested a method that
hardware layout. This structure was time-effective and robust relies on SG energy privacy protection. By analyzing a
enough to endure a detection accuracy of 95%. It was also long-term trend, they were able to utilize CNNs to spot
confirmed that it detects energy wasted due to customers’ lack any strange behavior in the metering data. In addition, the
of morals and circumventing of regulations, both of which paillier algorithm was used to ensure the confidentiality
are notoriously hard to track down using more conventional of transmitted energy data. This solution demonstrated the
methods of investigation. concomitant success of data privacy and authentication. The
A newly proposed detection method, electricity theft experimental results showed that the modified CNN model
detection using deep bidirectional RNN (ETD-DBRNN), was able to identify aberrant behaviors with an accuracy of
which is used to capture the internal characteristics and the up to 92.67 % [191]. Using the new method of multiple
external association by analyzing the energy consumption pricing (MP), [192] demonstrated a method of stealthy power
records, thereby overcoming the shortcomings of existing theft (HET). As a means of building the HET assault,
ML based detection methods is presented in [189]. Validation they suggested an optimization problem with the goal of
of this strategy was shown by experiments using real-world maximizing attack revenues while evading existing detection
data. mechanisms. Two algorithms intended to hack smart meters
It was observed that this technique better captures the were also developed. In order to demonstrate the potential
information of the power use records and the intrinsic of HET attacks, the authors discovered and exploited a wide
characteristics between normal and abnormal electricity variety of new vulnerabilities in smart meters. To defend
usage patterns than the currently available methods. In reality, SGs against HET attacks, the authors recommend a number
energy thieves’ meter readings should have a stronger of defensive and detection techniques, including restricting
correlation to the value of power theft loss than the readings of the attack cycle, selective protection on smart meters,
honest users. Due to this, [190] systematically formulated the and an updated billing methodology. The recommended
issue of power theft identification as a time-series correlation countermeasures successfully reduced the attack’s effect with
analysis problem. Two coefficients were constructed to little effort or expense.

1488 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

FIGURE 7. Possible energy theft techniques.

The reintroduction of old problems like energy theft years has created a large quantity of data (e.g., price-
necessitated the development of cutting-edge detection based, incentive-based and environmental-based. In this
systems and architecture predicated on data analysis, machine part, we will discuss the IA methods used in EMS on
learning, and predictive modelling made possible by the the building level. By using AI techniques, we can plan
advent of digital power meters. To identify electricity meter for and control numerous energy assets automatically via
manipulation in its early stages, [193] exhibited a multidi- the EMS, therefore addressing a number of problems
mensional innovative detection technique and architecture plaguing the field of energy management. There has been
by comparing a collection of energy demand time series. a noticeable growth in the use of AI techniques in DR
Newer monitoring systems can often only analyze one programs since 2013. Price-based programs and domestic
time series; thus, this approach was a welcome addition consumer types have seen the most use of these AI tech-
and improvement. Their goal was to find ways to spot nologies, followed by small-scale industrial and commercial
electrical theft, and they offered three different approaches structures [8].
to preprocessing data to do so. This approach demonstrated Numerous articles discuss the use of AI in energy DR
the metric’s robustness in the face of tampered data. With initiatives. For a broader view, see [202], which explores
detection rates of over 90%, the authors demonstrated the the current status of DR applications and analyses the AI
primary advantage of combining numerous data sources at methodologies used across a variety of DR scheme types and
once, rather than relying on each one of them separately. customer types. In addition, a comprehensive overview of
This approach also demonstrated the usefulness of comparing corporations, innovators, and European-funded commercial
data from several houses without first categorizing them into initiatives using AI for DR is offered. In particular, [194]
comparable groups. examines the current AI-based approaches to cloud EMS
and how blockchain technology may be incorporated into
F. ENERGY MANAGEMENT SYSTEM them. However, there is a significant number of unanswered
In order to improve building energy efficiency and DR questions about blockchain that need be investigated in the
programs, the fast deployment of smart meters in recent next years, including its expensive development and storage

VOLUME 11, 2023 1489

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

costs, the absence of uniformity in the field, and the scarcity Only a small number of European nations have so
of specialized knowledge. far allowed aggregators and DR access to their market-
The research examples below make use of Supervised places [201]. In Europe and the US, DR is seen as a crucial
Learning and Unsupervised Learning approaches. In [195], instrument for combining renewable energy sources with
an MLP-based deep learning model is employed to improve a consistent electrical supply [202], [203]. Similar issues,
load consumption and storage management in light of such a restrictive power market or a monopoly, are, however,
variable pricing. By optimizing the scheduling of domestic virtually nonexistent in many developing nations [204]. One
appliances and RES production, [196] a deep ANN and aspect influencing the creation and use of such a DR program
Genetic Algorithm help to lower energy consumption during is the nation in which it is implemented. The main reason
peak hours. Loads for EMS may be identified by means for this is that every nation has unique energy resources,
of smart plugs using supervised learning algorithms like regulatory and policy frameworks, and power markets. As a
DT and Naive Bayes. Using regression trees and RF, the result, the supply side needs for DR projects change. For
authors of [197] build a model of a heating system with instance, solar energy is more relevant in some nations than
the goal of controlling its operation. In [106], authors use wind energy is in others, necessitating a different change in
an ANN-based stable price prediction model to address the demand [205]. Numerous countries across the globe, such as
problem of pricing uncertainty in EMS. After a demand Germany, Denmark, Finland, and others, have considerable
response (DR) event mandated by the power company to levels of varying renewable energy in their power network
minimize peak usage, [198] creates a household scheduling and are battling various problems due to fluctuation while
controller utilizing the hybrid lightning search algorithm integrating at the grid station due to fluctuations and flickers
ANN to forecast the best ON/OFF state for home electrical which causes reduction in the stability of the grid [206].
equipment. While [199] provides a prediction approach based Programs across the European countries are heavily depends
on LSTM of the end-user reaction behavior to incentive- on initiatives taken by each country its unique legislative
based DR program, [62] employs ANN to anticipate and environment because the unified European electricity market
plan building appliances energy consumption and genetic does not yet exist [207]. Cross-border connectivity would
algorithms for job scheduling. entail the transmission of power via high-voltage lines
The articles under consideration in this area have one major between nations, but it would need the right infrastructure
flaw: they depend on the end user being completely familiar and transactional procedures [208]. Any nation that has one or
with the surrounding environment. Although price-based DR more network points is in charge of these connecting points.
programs get the lion’s share of attention in the literature, The balance-responsible party (BRP) must create scheme
it is clear that a variety of incentive-based DR methods would for power generation and load balanced system which is
benefit the distribution network greatly. To yet, only a few of known as the balancing duty [200]. Day-ahead, capacity, next
papers have shown even somewhat accurate models of the day, intraday, and balancing energy markets are currently
appliances that can be controlled by EMS. The RL technique available in each individual EU member state. Each of these
does not need any knowledge of the system model, unlike markets has its own participation requirements [207].
standard model-based approaches. Table 10 summarizes this The British Empire (UK): The UK’s Office of Gas and
whole process by dividing each source into three distinct Electricity Markets (OFGEM) was the first organization
categories: data-driven method, DR program, and client type. which is allowed to initiate a capacity market in 2013 when
Energy Act was passed to encourage the development of
DR [209]. The US began testing DR in the early 2000s.
V. DR PROGRAMS EMPLOYED IN DIFFERENT COUNTRIES UK was one of the first market to implement DR in their
The present emphasis in Europe is on DR monitoring and power sectors and energy markets [210]. Short term operation
SG-enabled monitoring systems, according to a study of service (STOR) like other initiatives were taken in order to
scientific and applied sources. For the past 20 years, make a balance between supply and demand [211], the UK
SG initiatives have been carried out; however, because of market for DR in the electricity sector began to take off.
inconsistencies in implementation of DR in EU, current In addition to Belgium and Switzerland, the DR analysis
efforts may be difficult in case of transferring smart grid reveals that the UK was the leader among the European
information data and across the border with nearby countries. nations that would be the most useful for comparison [23].
Another issue is the lack of an EU-wide program or The commercial and public sectors make up around one-third
unified DR policy. Countries that create SG-based power of the portfolios of aggregators, and the UK is one of the best
supply networks can let their electrical users take part in nations to implement these schemes and participant got fully
DSM initiatives. Source [200] reports that while domestic interested in participating more and more. [210].
appliances and commercial services shares for 30.9% and Germany: Germany is frequently seen as a pioneer in the
30.4%, respectively, of the total electricity demand in Europe, management of renewable energy [211] which separated its
industrial users account for 36.1% of that demand. Previous activities to liberalize its power sector. Additionally, it has the
studies have shown that DSM is most advantageous when most trustworthy system available, ENTSO-E [200]. Energy-
applied to larger consumption area as compared to lower one. saving initiatives are far more prevalent in Germany than

1490 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

TABLE 1. Different forecasting models and evaluation metrices for DR programs using AI techniques.

VOLUME 11, 2023 1491

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

TABLE 2. (Continued.) Different forecasting models and evaluation metrices for DR programs using AI techniques.

DR [202]. The German demand reduction control is playing DR system [214], [215], [216]. The network’s functionality is
a key role in reducing imbalance of supply because it is based managed by Helsinki, Finland-based transmission company
on broad, high-demand targets for the medium and long term, Fingrid Ltd., which also responds to customer demand
out of which few are longer-term and more ambitious than using the market’s established processes [217], [218]. Active
those outlined in the EU energy efficiency regulation [211]. retailers typically insistently offer price tariffs to end-
The penetration of RES into the energy system and the users when the nation moves to commercial activities,
dependability of the grid have a greater influence on the prompting a change in consumption based on the market
German DR’s value [202]. The wholesale market in Germany environment [219].
has become too volatile because to the significant penetration Switzerland: Typically, local governments design and
of RES [212]. The technological capacity of German DR administer DSM policies. Swiss utilities have been using
is 6.4 GW/h, from which 3.5 GW/h is attained within the ToU and ripple control, two DSM technologies, longer than
present market and regulatory structure. DR might be as other European nations. To maintain the stability of the
high as 10 GW [213] if a better market and regulatory electricity grid, fluctuation control is a common method of
environment are created. The nation requires a reliable system load control. The conventional power signal is switched to
for establishing prices [211]. a higher frequency signal (50 Hz). Loads such as heaters,
Finland: Finland is regarded as an innovator in the use of electric boilers, and public street lighting can all be turned
smart meters and a leader among European nations with a on and off.

1492 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

FIGURE 8. IoE sources with big data for energy markets at different level (generation, transmission, and distribution) at different consumer level
(residential, commercial and industrial) for different function blocks (cloud, server, platform and HMI application) using different hierarchical network
(HAN, WAN, NAN).

Belgium. DR projects are receiving more attention, partic- retirement of some nuclear power reactors and the constant
ularly in the residential sector, as a result of the scheduled development in RES capacity [40]. In order to encourage

VOLUME 11, 2023 1493

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

FIGURE 9. Different IOE objectives.

flexibility, the national transmission operator is searching for A DR aggregation pilot program has been running in Estonia
measures to reduce energy costs. Such flexibility is especially since the end of 2017 [224]. The main barrier for aggregators
crucial during the winter, when demand for electricity peaks in Latvia at the moment is the lack of a legal framework
at around 12,000–14,000 MW, or around 2000 MW more than defining the duties and responsibilities of aggregators as well
in the summer [220]. as the compensation mechanisms amongst various energy
Baltic States. The balancing management of the Baltic system participants [221]. The Estonian start-up Fusebox and
power system is become more difficult for two reasons. First, the Lithuanian transmission firm ‘‘Ignitis’’ began working
similar to trends in Central and South-Eastern Europe, the together in 2020 with the goal of creating Lithuania’s first
power of easily regulated traditional major power plants is independent power demand aggregator. In terms of energy
declining in the Baltics, while the output of less regulated, security and efficiency, the power demand service is a recent
less foreseeable, and distributed sources, such as wind power, development in the Lithuanian electricity market.
is increasing [221]. Second, by 2025, the Baltic states intend
to cut their ties to Russia’s single power grid, forcing the VI. BLOCKCHAIN BASED DEMAND RESPONSE IN SMART
creation of additional flexibility sources to maintain electrical GRID
balance under both normal and unusual circumstances [222]. The production of power and economic losses of generated
This will be made possible by a project that was endorsed power must be reduced for which one strategy is to use
by the Commission of the Connecting Europe Facility of smart energy solutions (SES) more frequently. However, the
the European Union Infrastructure Network, which aims to dissemination of SES should be promoted as a tool to achieve
make the Baltic power systems independent with regard to shared objectives rather than for its own sake [225]. Artificial
frequency with Poland and other countries in continental intelligence (AI) and machine learning (ML) are appeared
Europe. Due to the underdevelopment of the energy-intensive as one of the best recently emerged critical technologies
industry in the Baltics, the DR potential is restricted to for supporting DR. Due to the complex nature of DR
smaller consumer markets [223]. Consumers in Latvia and operations, exploitation of bid data and the demand for near
Estonia can now access implicit DR thanks to energy supply actual decisions the internet of energy (IoE), which uses the
agreements where the retail price is linked to the spot price. networks to integrate and communicate different transducers

1494 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

FIGURE 10. Centralized classification of different issues associated with IoE.

and sensors deployed for different measurements, integration use of big data processing, ubiquitous computing, and M2M
of RES into smart grids, and several other techniques communication [226]. The IoT extends the reach of the
plays its crucial role and transformed power production internet to include the devices deployed on the energy system
and supply in to new era of technological development by by using standardized communication protocols [227]. IoE
implementing proper schemes. DSM programs are made up is characterized by a variety of energy sources, supply and
of energy saving and efficiency improvement consideration, demand coordination, centralization and decentralization,
DR programs, and domestic or commercial load monitoring and extensive public involvement [228]. IoE promises a
programs. By changing the consumer electricity usage number of significant advantages, like power monitoring,
schedule, smart buildings can improve the adaptability of the energy demand management, expanding renewable energy
power load and provide excellent opportunities for power integration, less wasted energy, fewer power failures, self-
DR’s. However, there hasn’t been much research done in regulation, and resource management. In order to accomplish
this area, so it’s unclear how big the potential effects of efficient, clean, and secure energy consumption, the Internet
smart houses engaging in power demand response would of Energy (IoE) is a grid that connects numerous distributed
be. Smart houses in particular can take part in power DR power harvesting technologies, electrical energy storage
in two different ways: (a) By spreading out the load across devices, and various types of loads [24]. To accomplish
a specific time period, smart homes can flatten the load efficient, clean, and secure energy consumption, a variety of
curve by ‘‘peak shaving and valley filling.’’ And (b) To loads are used [228].
reduce residential power costs, smart houses can move a Smart Grids with IoE. In order to I lower electricity
load to a time period that is more affordable. Internet of prices, (ii) control peak loads, and (iii) lower electricity costs
Everything (IoE) is a term that emerged from the internet for RES generation variations, which may disrupt energy
of things (IoT) to describe industrial applications that make systems [229], smart grids allow a two-way link between

VOLUME 11, 2023 1495

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

customers and operators. Contrarily, an SG facilitates a solu- systems, PV modules may be tracked and controlled just like
tion for energy production, supply, and storage, with the most any other node in the Internet of Things [235].
up-to-date information on energy pricing attached [230]. Distribution system operators (DSOs) are anticipated to
When conventional methods of grid fortification are either focus on the effect of PV intermittency on optimal power
impractical or too costly to implement, SGs in distribution flow (OPF) and voltage regulation as solar power becomes
networks need creative approaches to circumventing network more widely used. Using DR capabilities to their full
restrictions [231]. Smart metering’s widespread use and potential is one effective technique for overcoming such PV
the potential to combine several decentralized small-scale integration challenges [242]. Technology developments have
RES [34] pose a significant threat to the limitations of also allowed for a wider range of inverter control tactics to
traditional centralized power networks. These facts point to be included into systems. In addition, these inverters carry
the need of a shift toward a decentralized and distributed out a number of functions, including as regulating voltage,
energy system [230]. However, the high penetration rate and generating active and reactive power, and extracting energy
fast pace of variable RES and battery energy storage systems from a photovoltaic (PV) module or array [243]. In terms
(BESS) make energy management in decentralized energy of network reliability, frequency is considered crucial [234].
systems challenging [232]. If the frequency of an object deviates from the usual, it means
However, we will discuss the major challenges that DR that either more power is needed or there is too much
systems provide in the area of Smart Grids. The first already [234]. Since the system frequency decreases when
source of power market interoperability is the plethora of demand exceeds supply and increases when supply exceeds
utility providers, retailers, and hardware/software developers. demand, DR is a useful tool for keeping the network’s
Second, the number of agents and the complexity of their frequency stable [235]. Most DR equipment has fast on/off
behaviors in power systems are both increasing due to controls to provide a steady power supply from the consumer
widespread use of smart meters, IoT devices, and DERs. end [81].
Finally, a decentralized architecture that at least offers the In a dynamic distribution system, single-phase PV invert-
potential for a settlement agreement in energy transactions ers may introduce power inconsistencies from dispersed
is required to entice buyers and diverse actors in the energy generators. Single-phase devices such inverters, voltage
sector to engage in large-scale DR schemes and assist market regulators (VRs), and capacitor banks (CBs) may break the
expansion [233]. To capitalize on the expanding market for three-phase symmetry of voltages and currents when voltage
energy and the expanding set of stakeholders, businesses control and DR are employed together, leading to an increase
are turning to business-to-business (B2B) e-commerce plat- in imbalance levels. Distributed photovoltaic systems may be
forms. Although standardized market processes serve as connected into the power grid using either a smart PV inverter
the backbone of such platforms, they may be modified or a regular PV inverter. Conventional PV inverters are unable
to better serve their intended purpose. The provision of of performing the complex control tasks required by PV
energy and supplementary services are only two examples systems (such as real power restriction, fixed power factor
of the kinds of significant tasks that may be balanced via regulation, volt-var control, volt-watt control, and frequency-
market procedures. Providers may also provide ancillary watt control), while smart PV inverters are capable of doing
services, which are customized solutions for problems like so. Therefore, smart PV inverters may reduce the number of
emergency power supply or energy conservation. An operator voltage and frequency control devices required in an electric
of the market commences the delivery of balancing services power grid, which in turn reduces the cost of both of these
in accordance with internal/external norms and generator factors [244].
schedules [234]. Technique of conveying messages. Recent advances in
The Internet of Things with renewable power sources like communication technology [89] have made it possible for
the sun. Solar photovoltaic (PV) farms have had access to AI modern SGs to transfer data and information swiftly and
methods for improving modelling, operational management, reliably in both directions. To increase power dependability
and output forecasting for over two decades [235]. Every and quality and stop electrical blackouts, DR’s marketing
DR system consists of two main parts: the controller and the and emergency signals may be sent utilizing a two-way
electrical appliance. To rephrase, the DR program helps to communication system (wireless, wire, GSM, and the
maintain a healthy equilibrium between power manufacturing internet). Utility billing disputes may be resolved faster with
and utilization, and the two control loops (additional control the use of this technology, and customers who verify DRs can
and DR control) collaborate to maintain order in the get incentive offers. Matching real-time supply and demand
system [236]. By facilitating real-time data sharing from data, integrating, and dispersing demand, and generating
PV sensors and allowing for remote controllability over the energy transactions and DR are all possible thanks to the
operation of solar units, IoT may aid in the identification IoE. To save on transmission costs, energy data is stored
of breakdowns and defects, as well as the performance of locally or on a network at a designated node [236]. Through
predictive and preventive maintenance [237]. The Internet of these protocols, the many nodes may exchange information
Things (IoT)-based smart PV monitoring systems are given with one another and the central control or decision nodes.
in [238], [239], [240], and [241]. Using smart monitoring Technologies such as LoRa or Sigfox (which provides the

1496 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

foundation for cloud-based services in future grids), ZigBee, individual operations and services. With more and more
Z-Wave, Bluetooth, Wi-Fi, and cellular technology such as vulnerabilities caused by the centralization of IoE, the
LTE-4G and 5G networks are only a few examples [237]. adoption of innovative solutions is increasing rapidly. The use
Problems with data transfer rates, communication delays, of smart contracts and encryption in blockchain technology
security, and device connection may all be mitigated with serves as an example of how decentralization and autonomy
the use of 5G technology [239]. 5G’s reduced latency and may be achieved. [245] As renewable energy prices decrease
high reliability make it a practical replacement for hardwired and new technologies become more affordable, consumers
connections. When it comes to SG communication, it’s rather are demanding a more efficient, greener, and long-term
uncommon to have stringent performance requirements, such energy system. Through its decentralized trading mechanism,
as extremely low latency (usually less than 1ms) [240]. blockchain technology will promote sustainable electricity
As 5G technologies continue to evolve, a demonstrably better use and help bring about a circular economy [246]. The
DR infrastructure will emerge [239], raising the standard in blockchain technology was first developed for the digital
transmission, reliability, safety, and connection. The latest currency Bitcoin, and it guarantees secure transactions
standard for communicating DR signals via IP networks by connecting buyers and sellers directly. When Bitcoin
(like the internet) is called Open automated demand response was successful, the next generation of blockchain tech-
(OpenADR v2.0). Smart meters, internet-connected devices, nology, the smart contract platform, was developed [233].
and distributed energy resources (DERs) have led to a A new technology built on the same Bitcoin-inspired
growth in the number of agents and the complexity of premise, blockchain for smart contracts allows the requisite
activities associated to power systems [233]. For instance, distributed-operations-enabling techniques via its decentral-
the Enel Info +, Smart Demo Grid, and FLEXICIENCY ized network architecture [13]. Power providers may utilize
projects have all built new smart meters with a dedicated blockchain for smart contracts on DERs, such as to accept,
communication channel at home area networks [239]. The fund, and sell electricity generated by renewable energy
need for outsourcing platforms is rising at the present time. sources (RES) [233]. Due to its importance in facilitating
When it comes to customer service, several platforms use the implementation and usage of RES and the smart grid,
employees to reduce customers’ wait times [240]. Each DR blockchain technology is poised to play a pivotal role in the
installation must have online bidirectional connection with expansion of the IoE market. Secure and efficient energy
a national control center [239], and DR aggregators may transactions are made possible by the widespread use of
employ new technologies to make it possible. blockchain technology and the Internet of Things (IoT).
Issues with cyber security in smart grids. The use of Simultaneously, it enables the integration of electricity from
smart grid technology to handle DERs and EVs necessitates distributed generation, utility grids, and other sources [79].
enhanced sensing, communication, and control mecha- More work has to be done to establish potentially effective
nisms [241]. Infrastructure failures may occur as a conse- system protocols for blockchain-based power systems, and
quence of cyber assaults, cascade failures, blackouts, and all promotional opportunities should be taken advantage
other forms of attack on the electric grid [242]. In a residential of [240].
home management system, for instance, the usage of heater The analysis of current research that uses distinct indices
or air conditioner data throughout the summer and winter to decrease the effect of DR technology on consumers reveals
seasons might provide insight about inhabitants’ availability. a number of particularly noteworthy studies. By generating
The burglars might use this knowledge to plan an attack on the consumer convenience and demand rebound indices and
home. Because the regulators may access the central server developing objective functions based on these indices, the
at any time, which has all the information and data required authors of [247] provide a novel approach to lowering cus-
from the utility suppliers, this is an extremely important tomer dissatisfaction, system capacity, and demand rebound.
problem to solve [243]. Smart grids have challenges from In [248], the authors propose a new linear discomfort index
the huge variety of devices connected through wide area (DI) formulation that takes into account user preferences
networks. Securing isolated devices within the context of a during normal appliance usage as part of a self-scheduling
larger network is the biggest challenge [244]. More thought model for HEMS. Building a sustainable energy infras-
has been given to cyber security in the context of modern DR tructure requires a plethora of different strategies. As an
initiatives and the corresponding communication networks, alternative, the transition should be carried out carefully,
such as OpenADR in the United States and China’s reform with consideration given to all relevant technologies and
of the management system for air conditioning in public approaches.
buildings [239].
Protecting the smart grid using blockchain technology. VII. CHALLENGES AND OPPORTUNITIES OF USING AI IN
The goal of creating a decentralized energy system using DEMAND RESPONSE
micro-grids, RES, and EVs necessitates a trading model that Artificial neural networks are one of the most widely
is robust, extensible, and effective. In addition, we need to used families of techniques and have mostly been used in
upgrade our communication, transmission, and distribution forecasting applications. The researchers have implemented
networks so that we may take control of our networks’ ANNs utilizing a single hidden layer as well as ‘‘deeper’’

VOLUME 11, 2023 1497

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

multi-layer designs for both load and price forecasting. convergence rates, also GAs occasionally use sophisticated,
In contrast hand, the collection of chosen variables that will occasionally unintuitive functions in selection and crossover
be used as inputs, the learning algorithm, and the optimization operators and, if not correctly adjusted, can suffer from early
of their hyperparameters can all have a significant impact convergence and unpredictable results [40].
on how well they function, and there is no one approach Key obstacles in allowing AI approaches are the availabil-
that can ensure the best selection of these. Additionally, ity and access to high-quality data sources. There is a lack of
ANNs can be resource intensive and typically need a lot high-quality data in the energy industry, making it difficult
of data in order to perform better than other less flexible to create robust systems [10]. In addition, there is a need to
methods. Due to the current low adoption of DR programs, enhance the freshness, integrity, correctness, and consistency
this may be a challenge for DR applications. Compared of data used in energy AI applications [9]. Data management
to ANNs, supervised machine learning approaches are less and data governance techniques will become more important
flexible, greater bias procedures, and primarily rely on feature as digital technologies advance so that these issues may be
engineering and selection to achieve effective results. On the dealt with effectively.
other hand, supervised techniques like gradient boosting Widespread adoption inside and across applications
and regression trees [249], [250] can manage missing requires not just agreed upon methods, but also the accessibil-
data effectively than ANNs. The benefit of AI forecasting ity of data points and computation solutions. Companies are
algorithms to produce forecasts that span several horizons in wary of sharing data to protect proprietary information and
time and space, as well as the ability to include uncertainty keep their competitive edge. Distribution System Operators
in the forecasts, results in predictions that are more useful. (DSOs) and Transmission System Operators (TSOs) may
The performance of AI techniques for predicting, on the other benefit greatly from an open and honest interchange of
hand, can vary depending on the hyper-parameter tuning and operational information from distribution and transmission
feature engineering they employ. grids. Because of this, TSOs and DSOs must establish what
There are not so much labelled data points available in data they need, the data’s quality, the data’s owner, and
the current DR environment to categorize customers [251]. how confidentiality and openness may be maintained [8].
As a result, the only workable method to tackle the However, optimization schemes and solutions can’t be
issue of segmenting electrical customers is to use unsuper- created or tested without access to public data sets. Green
vised/clustering models. Although clustering algorithms are Button and OpenEI in the United States and the ENTSO-E
useful in this application, there are several difficulties with Integrity Platform in the European Union are only two
them. These algorithms, among others, suffer from the ‘‘curse examples of the many projects that encourage energy data
of dimensionality,’’ require data pre-processing to function, exchange among stakeholders. However, there is a need for
and it is extremely difficult to evaluate the results [34]. a rise in publicly accessible data on energy, since doing
Reinforcement Learning (RL) techniques typically do not so, together with opening up data from the public sector
call for the use of an environment model, in contrast to more and adopting common data standards, may assist to spur
conventional DR control mechanisms like Model Predictive innovation.
Control (MPC) [35]. This gives designers of DR control Data mining, machine learning, data analysis, data pro-
systems that consider consumer preferences an advantage. cessing, and data visualization are just some of the cutting-
Deep RL has also been demonstrated to perform better edge methods that have been used to the energy industry.
in high-dimensional problems [35]. The design of reward The utilization of Big Data has become more simpler
signals, however, is a major problem for RL generally, with because to the sophisticated technologies that have been
consequences for its proper application in DR [35]. There steadily improving and are now more widely available. The
have been many instances where RL agents have discovered development of new enterprises and the provision of new
novel ways to manipulate their environments so that rewards services is facilitated by these cutting-edge data analysis
are delivered but with unfavorable rules [35]. To the best of methods. For this reason, it is essential to discover untapped
our knowledge, there is very little research literature on this markets by analyzing current data and developing data-
subject in the energy DR. driven business models examines forty data-driven energy
Mostly, nature inspired algorithms are used for schedul- industry startups as case studies of these novel business
ing purpose. The scheduling problem, in general, can be models.
extremely difficult, non-linear, and non-convex. Due to their
capacity for exploration and exploitation, these group of VIII. CONCLUSION
algorithms can locate interesting solutions in a fair amount of New demands, such as electric vehicles, heat pumps,
time [40]. They are parallel algorithms, which are robust and and the rising penetration of distributed energy resources
adaptable to changing conditions and environments, among (DERs) provide new difficulties for electrical networks.
other major benefits [40]. On the other hand, nature-inspired Grid operators may keep the electrical grid in balance by
approaches do not promise to discover the best solution, investing low-cost on implementing proper DR schemes,
and certain algorithms have their own limitations. Like PSO, while eliminating the need for expensive enhancements
which can become stuck in local optima and have slow of the electrical network or investing in a great deal of

1498 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

expensive back-up generation, so DR providing a cost- REFERENCES

effective response to these difficulties. There is a significant [1] C. Eid, P. Codani, Y. Perez, J. Reneses, and R. Hakvoort, ‘‘Managing
push to add domestic and commercial customers into the DR electric flexibility from distributed energy resources: A review of
incentives for market design,’’ Renew. Sustain. Energy Rev., vol. 64,
portfolio, despite the fact that DR programs were initially pp. 237–247, Oct. 2016.
aimed for a selected group of major industrial and heavy [2] X. Luo, J. Wang, M. Dooner, and J. Clarke, ‘‘Overview of current
commercial users. This shift demands to appropriately choose development in electrical energy storage technologies and the appli-
cation potential in power system operation,’’ Appl. Energy, vol. 137,
the end users contributing to a certain consumption hour, pp. 511–536, Jan. 2015.
but also to plan their usage, control units for DR, and [3] M. Andoni, V. Robu, D. Flynn, S. Abram, D. Geach, D. Jenkins,
set the reward/ punishment schemes. Artificial intelligence P. McCallum, and A. Peacock, ‘‘Blockchain technology in the energy
sector: A systematic review of challenges and opportunities,’’ Renew.
(AI) solutions have been widely employed by researchers Sustain. Energy Rev., vol. 100, pp. 143–174, Feb. 2019.
to accomplish these goals, particularly in cases when [4] G. Bedi, G. K. Venayagamoorthy, R. Singh, R. R. Brooks, and
more conventional methods failed to provide acceptable K.-C. Wang, ‘‘Review of Internet of Things (IoT) in electric power and
energy systems,’’ IEEE Internet Things J., vol. 5, no. 2, pp. 847–870,
outcomes. Apr. 2018.
To identify and explore the trends for AI techniques in the [5] S. Zhou and M. A. Brown, ‘‘Smart meter deployment in Europe: A
energy DR industry, the authors here investigated through comparative case study on the impacts of national policy schemes,’’
more than 150 articles, as well as 35 plus businesses and J. Cleaner Prod., vol. 144, pp. 22–32, Feb. 2017.
[6] Office of Gas and Electricity Markets (OFGEM), Transition to Smart
commercial efforts, and 20 significant projects. According to Meters. Accessed: Dec. 2, 2018. [Online]. Available: https://www.ofgem.
the research that was analyzed for this paper, AI methods are gov.uk/gas/retail-market/metering/transition-smart-meters
a potential tool for DR. Artificial intelligence (AI) usage is [7] S. Howell, Y. Rezgui, J.-L. Hippolyte, B. Jayan, and H. Li, ‘‘Towards
the next generation of smart grids: Semantic and holonic multi-agent
crucial to the future success of DR programs. It is crucial for management of distributed energy resources,’’ Renew. Sustain. Energy
successful use in a DR context to have a better grasp of AI Rev., vol. 77, pp. 193–214, Sep. 2017.
methodologies and their limits. [8] I. Antonopoulos, V. Robu, B. Couraud, D. Kirli, S. Norbu, A. Kiprakis,
D. Flynn, S. Elizondo-Gonzalez, and S. Wattam, ‘‘Artificial intelligence
Our analysis revealed that many distinct AI methods are and machine learning approaches to energy demand-side response: A
in use, but it is evident that certain methods are better suited systematic review,’’ Renew. Sustain. Energy Rev., vol. 130, Sep. 2020,
than others for certain jobs. Using supervised learning, it was Art. no. 109899.
[9] H. T. Haider, O. H. See, and W. Elmenreich, ‘‘A review of residential
shown that ANNs are widely utilized for short-term load and demand response of smart grid,’’ Renew. Sustain. Energy Rev., vol. 59,
price forecasting. ANNs are often used for multi-variable pp. 166–178, Jun. 2016.
function approximation and regression. On the other hand, [10] N. Good, K. A. Ellis, and P. Mancarella, ‘‘Review and classification of
barriers and enablers of demand response in the smart grid,’’ Renew.
RL-based algorithms are typically employed to gather human Sustain. Energy Rev., vol. 72, pp. 57–72, May 2017.
input, making them well-suited for control tasks in HEMS [11] K. Dehghanpour and S. Afsharnia, ‘‘Electrical demand side contribution
that include a DR solution. For clustering problems at the to frequency control in power systems: A review on technical aspects,’’
Renew. Sustain. Energy Rev., vol. 41, pp. 1267–1276, Jan. 2015.
aggregator level, such as those involving DR customers,
[12] H. W. Pandey, R. Kumar, and R. K. Mandal, ‘‘Transformation of
unsupervised learning is often employed since it does Indian distribution sector: Opportunity and challenges for unlocking the
not need any previous knowledge of the categories. After demand response potential,’’ Renew. Energy Focus, vol. 42, pp. 221–235,
classifying and forecasting DR customers’ usage, aggregators Sep. 2022.
[13] M. Waseem, Z. Lin, Y. Ding, F. Wen, S. Liu, and I. Palu,
then set a date to activate DR participants and determine ‘‘Technologies and practical implementations of air-conditioner based
incentives and punishments. It has been noted that standard, demand response,’’ J. Mod. Power Syst. Clean Energy, vol. 9, no. 6,
stochastic optimization approaches are less accurate, and that pp. 1395–1413, 2021, doi: 10.35833/MPCE.2019.000449.
[14] H. Shareef, M. S. Ahmed, A. Mohamed, and E. A. Hassan, ‘‘Review
alternative, nature-inspired optimization techniques (such on home energy management system considering demand responses,
as swarm intelligence) may be required for certain jobs. smart technologies, and intelligent controllers,’’ IEEE Access, vol. 6,
The ideal pricing and scheduling method may also be pp. 24498–24509, 2018.
[15] I. Dusparic, A. Taylor, A. Marinescu, F. Golpayegani, and S. Clarke,
determined using multi-agent systems in game-theoretic ‘‘Residential demand response: Experimental evaluation and comparison
situations. of self-organizing techniques,’’ Renew. Sustain. Energy Rev., vol. 80,
Our findings also demonstrated that the industrial sector pp. 1528–1536, Dec. 2017.
[16] Y. Wang, Q. Chen, C. Kang, M. Zhang, K. Wang, and Y. Zhao, ‘‘Load
shares the increased enthusiasm of the research world for profiling and its application to demand response: A review,’’ Tsinghua
AI solutions in the DR sector, as seen by the proliferation Sci. Technol., vol. 20, no. 2, pp. 117–129, Apr. 2015.
of new start-ups in this area over the last several years. [17] J. R. Vázquez-Canteli and Z. Nagy, ‘‘Reinforcement learning for demand
response: A review of algorithms and modeling techniques,’’ Appl.
Although these AI-in-DR trends are well-established, further Energy, vol. 235, pp. 1072–1089, Feb. 2019.
study is obviously required to determine the best solutions for [18] Y. Wang, Q. Chen, T. Hong, and C. Kang, ‘‘Review of smart meter data
various situations. Many of the suggested remedies have not analytics: Applications, methodologies, and challenges,’’ IEEE Trans.
Smart Grid, vol. 10, no. 3, pp. 3125–3148, May 2019.
been subjected to rigorous testing in the form of large-scale,
[19] M. Waseem, Z. Lin, S. Liu, Z. Jinai, M. Rizwan, and I. A. Sajjad,
real-world trials and experiments. More research is needed, ‘‘Optimal BRA based electric demand prediction strategy considering
together with industry projects and large-scale testing, to pave instance-based learning of the forecast factors,’’ Int. Trans. Electr. Energy
the way for the development of more precise models and Syst., vol. 31, no. 9, Sep. 2021, Art. no. e12967.
[20] K. Zor, O. Timur, and A. Teke, ‘‘A state-of-the-art review of artificial
AI-based solutions. Taking this road will enable AI/ML intelligence techniques for short-term electric load forecasting,’’ in Proc.
methods to become commonplace in the energy DR industry. 6th Int. Youth Conf. Energy (IYCE), Jun. 2017, pp. 1–7.

VOLUME 11, 2023 1499

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

[21] G. H. Merabet, M. Essaaidi, H. Talei, M. R. Abid, N. Khalil, M. Madkour, [46] J. H. Holland, Adaptation in Natural and Artificial Systems: An Intro-
and D. Benhaddou, ‘‘Applications of multi-agent systems in smart grids: ductory Analysis With Applications to Biology, Control, and Artificial
A survey,’’ in Proc. Int. Conf. Multimedia Comput. Syst. (ICMCS), 2014, Intelligence. Cambridge, MA, USA: MIT Press, 1992.
pp. 1088–1094. [47] A. Mamun, I. Narayanan, D. Wang, A. Sivasubramaniam, and
[22] S. Lu, W. Gu, K. Meng, S. Yao, B. Liu, and Z. Y. Dong, ‘‘Thermal inertial H. K. Fathy, ‘‘Multi-objective optimization of demand response in a
aggregation model for integrated energy systems,’’ IEEE Trans. Power datacenter with lithium-ion battery storage,’’ J. Energy Storage, vol. 7,
Syst., vol. 35, no. 3, pp. 2374–2387, May 2020. pp. 258–269, Aug. 2016.
[23] J. Wang, H. Zhong, Z. Ma, Q. Xia, and C. Kang, ‘‘Review and prospect of [48] A. Darwish, ‘‘Bio-inspired computing: Algorithms review, deep analysis,
integrated demand response in the multi-energy system,’’ Appl. Energy, and the scope of applications,’’ Future Comput. Informat. J., vol. 3, no. 2,
vol. 202, pp. 772–782, Sep. 2017. pp. 231–246, Dec. 2018.
[24] U. B. Baloglu and Y. Demir, ‘‘A Bayesian game-theoretic demand [49] R. Eberhart and J. Kennedy, ‘‘Particle swarm optimization,’’ in Proc. Int.
response model for the smart grid,’’ Int. J. Smart Grid Clean Energy, Conf. Neural Netw., 1995, pp. 1942–1948.
[50] M. Dorigo and G. D. Caro, ‘‘Ant colony optimization: A new meta-
vol. 4, no. 2, pp. 132–138, Jul. 2015.
heuristic,’’ in Proc. Congr. Evol. Comput., Washington, DC, USA, 1999,
[25] P. Palensky and D. Dietrich, ‘‘Demand side management: Demand
pp. 1470–1477.
response, intelligent energy systems, and smart loads,’’ IEEE Trans. Ind.
[51] A. K. Kar, ‘‘Bio inspired computing-a review of algorithms and scope of
Informat., vol. 7, no. 3, pp. 381–388, Aug. 2011.
applications,’’ Expert Syst. Appl., vol. 59, pp. 20–32, 2016.
[26] S. Mohagheghi, J. Stoupis, Z. Wang, Z. Li, and H. Kazemzadeh, [52] A. Chakraborty and A. K. Kar, ‘‘Swarm intelligence: A review
‘‘Demand response architecture: Integration into the distribution man- of algorithms,’’ Nature-Inspired Computing and Optimization. 2017,
agement system,’’ in Proc. 1st IEEE Int. Conf. Smart Grid Commun., pp. 475–494.
Oct. 2010, pp. 501–506. [53] K. Lakshmaiah, S. M. Krishna, and B. E. Reddy, ‘‘An overview of
[27] C. Goldman, M. Reid, R. Levy, and A. Silverstein, ‘‘Coordi- bio-inspired computing,’’ in Proc. Int. Conf. Emerg. Res. Comput., Inf.,
nation of energy efficiency,’’ LBNL, Berkeley, CA, USA, Tech. Commun. Appl., 2018, pp. 481–492.
Rep. LBNL-3044E, Jan. 2010. [54] S. Mirjalili, S. M. Mirjalili, and A. Lewis, ‘‘Grey wolf optimizer,’’ Adv.
[28] M. H. Albadi and E. F. El-Saadany, ‘‘A summary of demand response Eng. Softw., vol. 69, no. 3, pp. 46–61, Mar. 2014.
in electricity markets,’’ Electr. Power Syst. Res., vol. 78, no. 11, [55] H. S. Bernardino and H. J. Barbosa, ‘‘Artificial immune systems for
pp. 1989–1996, Nov. 2008. optimization,’’ in Nature-Inspired Algorithms for Optimization. Berlin,
[29] S. J. Russell and P. Norvig, Artificial Intelligence a Modern Approach. Germany: Springer, 2009, pp. 389–411.
Upper Saddle River, NJ, USA: Prentice-Hall, 2010. [56] L. N. de Castro and J. Timmis, ‘‘An artificial immune network for
[30] G. James, D. Witten, T. Hastie, and R. Tibshirani, An Introduction to multimodal function optimization,’’ in Proc. Congr. Evol. Comput., 2002,
Statistical Learning: With Applications in R (Springer Texts in Statistics). pp. 699–704.
New York, NY, USA: Springer, 2013. [57] N. U. Rehman, H. Rahim, A. Ahmad, Z. A. Khan, U. Qasim, and
[31] P. Norvig, ‘‘On Chomsky and the two cultures of statistical learning,’’ N. Javaid, ‘‘Heuristic algorithm based energy management system in
in Berechenbarkeit der Welt? Wiesbaden, Germany: Springer, 2017, smart grid,’’ in Proc. 10th Int. Conf. Complex, Intell., Softw. Intensive
pp. 61–83. Syst. (CISIS), Jul. 2016, pp. 396–402.
[58] O. I. Abiodun, A. Jantan, A. E. Omolara, K. V. Dada, N. A. Mohamed,
[32] Y. I. Cho and E. T. Matson, ‘‘Soft computing in artificial intelligence,’’
and H. Arshad, ‘‘State-of-the-art in artificial neural network applications:
in Advances in Intelligent Systems and Computing, vol. 270. Cham,
A survey,’’ Heliyon, vol. 4, no. 11, Nov. 2018, Art. no. e00938.
Switzerland: Springer, 2014.
[59] Y.-J. Kim, ‘‘Optimal price based demand response of HVAC systems in
[33] B. Buchanan, E. Feigenbaum, and J. Lederberg, ‘‘Heuristic DENDRAL: multizone office buildings considering thermal preferences of individual
A program for generating explanatory hypotheses in organic chemistry,’’ occupants buildings,’’ IEEE Trans. Ind. Informat., vol. 14, no. 11,
Mach Intell., vol. 4, p. 30, 1968. pp. 5060–5073, Nov. 2018.
[34] K. P. Murphy, Machine Learning: A Probabilistic Perspective. [60] T. Holtschneider and I. Erlich, ‘‘Optimization of electricity pricing con-
Cambridge, MA, USA: MIT Press, 2012. sidering neural network based model of consumers’ demand response,’’
[35] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction. in Proc. IEEE Comput. Intell. Appl. Smart Grid (CIASG), Apr. 2013,
Cambridge, MA, USA: MIT Press, 2018. pp. 154–160.
[36] R. Lu, S. H. Hong, X. Zhang, X. Ye, and W. S. Song, ‘‘A perspective [61] Y. LeCun, Y. Bengio, and G. Hinton, ‘‘Deep learning,’’ Nature, vol. 521,
on reinforcement learning in price-based demand response for smart no. 7553, pp. 436–444, 2015.
grid,’’ in Proc. Int. Conf. Comput. Sci. Comput. Intell. (CSCI), Dec. 2017, [62] D. Liu, Y. Sun, Y. Qu, B. Li, and Y. Xu, ‘‘Analysis and accurate prediction
pp. 1822–1823. of user’s response behavior in incentive-based demand response,’’ IEEE
[37] R. Lu, S. H. Hong, and X. Zhang, ‘‘A dynamic pricing demand response Access, vol. 7, pp. 3170–3180, 2018.
algorithm for smart grid: Reinforcement learning approach,’’ Appl. [63] M. S. Ahmed, A. Mohamed, H. Shareef, R. Z. Homod, and J. A. Ali,
Energy, vol. 220, pp. 220–230, Jun. 2018. ‘‘Artificial neural network based controller for home energy management
[38] R. Lu and S. H. Hong, ‘‘Incentive-based demand response for smart grid considering demand response events,’’ in Proc. Int. Conf. Adv. Electr.,
with reinforcement learning and deep neural network,’’ Appl. Energy, Electron. Syst. Eng. (ICAEES), Nov. 2016, pp. 506–509.
vol. 236, pp. 937–949, Feb. 2019. [64] Y. Wang, Q. Chen, D. Gan, J. Yang, D. S. Kirschen, and C. Kang, ‘‘Deep
[39] M. Babar, P. H. Nguyen, V. Cuk, and I. G. Kamphuis, ‘‘The development learning-based socio-demographic information identification from smart
of demand elasticity model for demand response in the retail market meter data,’’ IEEE Trans. Smart Grid, vol. 10, no. 3, pp. 2593–2602,
environment,’’ in Proc. IEEE Eindhoven PowerTech, Jun. 2015, pp. 1–6. May 2018.
[65] S. Ryu, H. Choi, H. Lee, H. Kim, and V. W. Wong, ‘‘Residential load
[40] Z. Beheshti and S. M. H. Shamsuddin, ‘‘A review of population-based
profile clustering via deep convolutional autoencoder,’’ in Proc. IEEE
meta-heuristic algorithms,’’ Int. J. Adv. Soft Comput. Appl., vol. 5, no. 1,
Int. Conf. Commun., Control, Comput. Technol. Smart Grids, Oct. 2018,
pp. 1–35, 2013.
pp. 1–6.
[41] X.-S. Yang, Nature-Inspired Optimization Algorithms. New York, NY, [66] G. Chalkiadakis, E. Elkind, and M. Wooldridge, ‘‘Computational aspects
USA: Academic, 2020. of cooperative game theory,’’ Synth. Lectures Artif. Intell. Mach. Learn.,
[42] K. O. Stanley and R. Miikkulainen, ‘‘Evolving neural networks through vol. 5, no. 6, pp. 1–168, Oct. 2011.
augmenting topologies,’’ Evol. Comput., vol. 10, no. 2, pp. 99–127, [67] K. Hayakawa, E. H. Gerding, S. Stein, and T. Shiga, ‘‘Online mechanisms
Jun. 2002. for charging electric vehicles in settings with varying marginal electricity
[43] J. Lehman, J. Chen, J. Clune, and K. O. Stanley, ‘‘Safe mutations for deep costs,’’ in Proc. 24th Int. Joint Conf. Artif. Intell., 2015, pp. 1–17.
and recurrent neural networks through output gradients,’’ in Proc. Genet. [68] H. Ma, D. C. Parkes, and V. Robu, ‘‘Generalizing demand response
Evol. Comput. Conf., Jul. 2018, pp. 117–124. through reward bidding,’’ in Proc. 16th Conf. Auto. Agents MultiAgent
[44] M. Mitchell, An Introduction to Genetic Algorithms. Cambridge, MA, Syst., 2017, pp. 60–68.
USA: MIT Press, 1998. [69] H. Ma, V. Robu, N. Li, and D. C. Parkes, ‘‘Incentivizing reliability in
[45] N. J. Vickers, ‘‘Animal communication: When I’m calling you, will you demand-side response,’’ in Proc. 25th Int. Joint Conf. Artif. Intell., 2016,
answer too?’’ Current Biol., vol. 27, no. 14, pp. R713–R715, Jul. 2017. pp. 1–8.

1500 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

[70] R. Kota, G. Chalkiadakis, V. Robu, A. Rogers, and N. R. Jennings, [92] C. Yang, Q. Cheng, P. Lai, J. Liu, and H. Guo, ‘‘Data-driven modeling
‘‘Cooperatives for demand side management,’’ in Proc. 7th Conf. for energy consumption estimation,’’ in Exergy for a Better Environment
Prestigious Appl. Intell. Syst. (PAIS @ ECAI), Montpellier, France, and Improved Sustainability 2. Cham, Switzerland: Springer, 2018,
Aug. 2012, pp. 969–974. pp. 1057–1068.
[71] P. Bajari, R. McMillan, and S. Tadelis, ‘‘Auctions versus negotiations in [93] G. M. U. Din, A. U. Mauthe, and A. K. Marnerides, ‘‘Appliance-level
procurement: An empirical analysis,’’ J. Law, Econ., Org., vol. 25, no. 2, short-term load forecasting using deep neural networks,’’ in Proc. Int.
pp. 372–399, Oct. 2009. Conf. Comput., Netw. Commun. (ICNC), Mar. 2018, pp. 53–57.
[72] J. Sun, E. Modiano, and L. Zheng, ‘‘Wireless channel allocation using [94] J. Jazaeri, T. Alpcan, R. Gordon, M. Brandao, T. Hoban, and C. Seeling,
an auction algorithm,’’ IEEE J. Sel. Areas Commun., vol. 24, no. 5, ‘‘Baseline methodologies for small scale residential demand response,’’
pp. 1085–1096, May 2006. in Proc. IEEE Innov. Smart Grid Technol. Asia (ISGT-Asia), Nov. 2016,
[73] N. Edalat, C.-K. Tham, and W. Xiao, ‘‘An auction-based strategy pp. 747–752.
for distributed task allocation in wireless sensor networks,’’ Comput. [95] S. Park, S. Ryu, Y. Choi, J. Kim, and H. Kim, ‘‘Data-driven baseline
Commun., vol. 35, no. 8, pp. 916–928, 2012. estimation of residential buildings for demand response,’’ Energies,
[74] W. H. El-Ashmawi, D. S. Abd Elminaam, A. M. Nabil, and E. Eldesouky, vol. 8, no. 9, pp. 10239–10259, Sep. 2015.
‘‘A chaotic owl search algorithm based bilateral negotiation model,’’ Ain [96] G. Escrivá-Escrivá, C. Álvarez-Bel, C. Roldán-Blay, and
Shams Eng. J., vol. 11, no. 4, pp. 1163–1178, Dec. 2020. M. Alcázar-Ortega, ‘‘New artificial neural network prediction method
[75] S. D. Ramchurn, P. Vytelingum, A. Rogers, and N. R. Jennings, for electrical consumption forecasting based on building end-uses,’’
‘‘Putting the ‘smarts’ into the smart grid: A grand challenge for artificial Energy Buildings, vol. 43, no. 11, pp. 3112–3119, 2011.
intelligence,’’ Commun. ACM, vol. 55, no. 4, pp. 86–97, Apr. 2012. [97] Y. Chen, P. Xu, Y. Chu, W. Li, Y. Wu, L. Ni, Y. Bao, and K. Wang, ‘‘Short-
[76] K. Kochanski, D. Rolnick, P. Donti, and L. Kaack, ‘‘Tackling climate term electrical load forecasting using the support vector regression (SVR)
change with machine learning,’’ in Proc. AGU Fall Meeting Abstr., model to calculate the demand response baseline for office buildings,’’
Dec. 2019, vol. 2019, p. GC33A-04. Appl. Energy, vol. 195, pp. 659–670, Jun. 2017.
[77] D. Zhou, M. Balandat, and C. Tomlin, ‘‘Residential demand response [98] A. Arunaun and W. Pora, ‘‘Baseline calculation of industrial factories
targeting using machine learning with observational data,’’ in Proc. IEEE for demand response application,’’ in Proc. IEEE Int. Conf. Consum.
55th Conf. Decis. Control (CDC), Dec. 2016, pp. 6663–6668. Electron. Asia (ICCE-Asia), Jun. 2018, pp. 206–212.
[78] D. Zhou, M. Balandat, and C. Tomlin, ‘‘A Bayesian perspective on [99] E. A. M. Klaassen, J. Frunt, and J. G. Slootweg, ‘‘Experimental validation
residential demand response using smart meter data,’’ in Proc. 54th of the demand response potential of residential heating systems,’’ in Proc.
Annu. Allerton Conf. Commun., Control, Comput. (Allerton), Sep. 2016, Power Syst. Comput. Conf. (PSCC), Jun. 2016, pp. 1–7.
pp. 1212–1219. [100] P. MacDougall, A. M. Kosek, H. Bindner, and G. Deconinck, ‘‘Applying
[79] C. M. Cheung, R. Kannan, and V. K. Prasanna, ‘‘Temporal ensemble machine learning techniques for forecasting flexibility of virtual power
learning of univariate methods for short term load forecasting,’’ in Proc. plants,’’ in Proc. IEEE Electr. Power Energy Conf. (EPEC), Oct. 2016,
IEEE Power Energy Soc. Innov. Smart Grid Technol. Conf. (ISGT), pp. 1–6.
Feb. 2018, pp. 1–5. [101] P. MacDougall, B. Ran, G. Huitema, and G. Deconinck, ‘‘Performance
[80] R. Pereira, J. Figueiredo, and J. C. Quadrado, ‘‘Computational models assessment of black box capacity forecasting for multi-market trade
development and demand response application for smart grids,’’ in Proc. application,’’ Energies, vol. 10, no. 10, p. 1673, Oct. 2017.
Doctoral Conf. Comput., Electr. Ind. Syst., 2016, pp. 323–339. [102] P. MacDougall, B. Ran, M. Klever, and G. Deconinck, ‘‘Value assessment
[81] Z. J. Liu, N. Xiao, X. Wang, and H. Xu, ‘‘Elman neural network model for of aggregated energy flexibility when traded on multiple markets,’’ in
short term load forecasting based on improved demand response factor,’’ Proc. 14th Int. Conf. Eur. Energy Market (EEM), Jun. 2017, pp. 1–6.
[103] P. MacDougall, B. Ran, G. B. Huitema, and G. Deconinck, ‘‘Predictive
in Proc. 7th Int. Conf. Power Electron. Syst. Appl. Smart Mobility, Power
control for multi-market trade of aggregated demand response using a
Transf. Secur. (PESA), Dec. 2017, pp. 1–5.
[82] J. Grant, M. Eltoukhy, and S. Asfour, ‘‘Short-term electrical peak black box approach,’’ in Proc. IEEE PES Innov. Smart Grid Technol.
demand forecasting in a large government building using artificial neural Conf. Eur. (ISGT-Europe), Oct. 2016, pp. 1–6.
[104] J. Ponoćko and J. V. Milanović, ‘‘Forecasting demand flexibility of
networks,’’ Energies, vol. 7, no. 4, pp. 1935–1953, Mar. 2014.
[83] J. Schachter and P. Mancarella, ‘‘A short-term load forecasting model aggregated residential load using smart meter data,’’ IEEE Trans. Power
for demand response applications,’’ in Proc. 11th Int. Conf. Eur. Energy Syst., vol. 33, no. 5, pp. 5446–5455, Jan. 2018.
[105] Z. Li, S. Wang, X. Zheng, F. de León, and T. Hong, ‘‘Dynamic demand
Market (EEM), May 2014, pp. 1–5.
response using customer coupons considering multiple load aggregators
[84] S. Takiyar, ‘‘Grid reliability enhancement by peak load forecasting with
to simultaneously achieve efficiency and fairness,’’ IEEE Trans. Smart
a PSO hybridized ANN model,’’ in Proc. 4th Int. Conf. Rel., Infocom
Grid, vol. 9, no. 4, pp. 3112–3121, Jul. 2018.
Technol. Optim. (ICRITO), Sep. 2015, pp. 1–6.
[106] R. Lu, S. H. Hong, and M. Yu, ‘‘Demand response for home energy
[85] S. M. Basnet, H. Aburub, and W. Jewell, ‘‘An artificial neural network-
management using reinforcement learning and artificial neural network,’’
based peak demand and system loss forecasting system and its effect on
IEEE Trans. Smart Grid, vol. 10, no. 6, pp. 6629–6639, Nov. 2019.
demand response programs,’’ in Proc. Clemson Univ. Power Syst. Conf. [107] G. Dutta and K. Mitra, ‘‘A literature review on dynamic pricing of
(PSC), Mar. 2016, pp. 1–5. electricity,’’ J. Oper. Res. Soc., vol. 68, no. 10, pp. 1131–1145, Oct. 2017.
[86] A. Jagannath and A. Amanna, ‘‘Realizing data driven and Hampel [108] A. Al-Mousa and A. Faza, ‘‘A fuzzy-based customer response prediction
preprocessor based adaptive filtering on a software defined radio testbed: model for a day-ahead dynamic pricing system,’’ Sustain. Cities Soc.,
A USRP case study,’’ in Proc. Int. Conf. Comput., Netw. Commun. vol. 44, pp. 265–274, Jan. 2019.
(ICNC), Mar. 2018, pp. 310–314. [109] H. Taherian, M. R. Aghaebrahimi, L. Baringo, and S. R. Goldani,
[87] S.-H. Lee and K.-I. Moon, ‘‘Forecasting and modeling of electricity ‘‘Optimal dynamic pricing for an electricity retailer in the price-
demand using NARX neural network in smart grid environment,’’ Int. responsive environment of smart grid,’’ Int. J. Electr. Power Energy Syst.,
Inf. Inst. Tokyo Inf., vol. 17, no. 12, p. 6439, 2014. vol. 130, Sep. 2021, Art. no. 107004.
[88] N. G. Paterakis, A. Taşcıkaraoglu, O. Erdinc, A. G. Bakirtzis, and [110] C. Huang, Y. Shen, Y. Chen, and H. Chen, ‘‘A novel hybrid deep neural
J. P. S. Catalao, ‘‘Assessment of demand-response-driven load pattern network model for short-term electricity price forecasting,’’ Int. J. Energy
elasticity using a combined approach for smart households,’’ IEEE Trans. Res., vol. 45, no. 2, pp. 2511–2532, Feb. 2021.
Ind. Informat., vol. 12, no. 4, pp. 1529–1539, Aug. 2016. [111] T. Ahmad, H. Zhang, and B. Yan, ‘‘A review on renewable energy and
[89] A. Singh, S. Vyas, and R. Kumar, ‘‘A preliminary study towards electricity requirement forecasting models for smart grid and buildings,’’
conceptualization and implementation of a load learning model for smart Sustain. Cities Soc., vol. 55, Apr. 2020, Art. no. 102052.
automated demand response,’’ in Proc. IEEE 7th Power India Int. Conf. [112] M. N. Akhter, S. Mekhilef, H. Mokhlis, and N. M. Shah, ‘‘Review on
(PIICON), Nov. 2016, pp. 1–6. forecasting of photovoltaic power generation based on machine learning
[90] N. G. Paterakis, J. P. S. Catalao, A. Tascikaraoglu, A. G. Bakirtzis, and and metaheuristic techniques,’’ IET Renew. Power Gener., vol. 13, no. 7,
O. Erdinc, ‘‘Demand response driven load pattern elasticity analysis for pp. 1009–1023, May 2019.
smart households,’’ in Proc. IEEE 5th Int. Conf. Power Eng., Energy [113] S. Sobri, S. Koohi-Kamali, and N. A. Rahim, ‘‘Solar photovoltaic
Electr. Drives (POWERENG), May 2015, pp. 399–404. generation forecasting methods: A review,’’ Energy Convers. Manage.,
[91] C. Ninagawa, S. Kondo, and J. Morikawa, ‘‘Prediction of aggregated vol. 156, pp. 459–497, Jan. 2018.
power curtailment of smart grid demand response of a large number of [114] A. Zendehboudi, M. A. Baseer, and R. Saidur, ‘‘Application of support
building air-conditioners,’’ in Proc. Int. Conf. Ind. Informat. Comput. Syst. vector machine models for forecasting solar and wind energy resources:
(CIICS), Mar. 2016, pp. 1–4. A review,’’ J. Cleaner Prod., vol. 199, pp. 272–285, Oct. 2018.

VOLUME 11, 2023 1501

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

[115] M. Sharifzadeh, A. Sikinioti-Lock, and N. Shah, ‘‘Machine-learning [137] P. Faria, Z. A. Vale, J. Soares, and J. Ferreira, ‘‘Particle swarm opti-
methods for integrated renewable power generation: A comparative study mization applied to integrated demand response resources scheduling,’’ in
of artificial neural networks, support vector regression, and Gaussian Proc. IEEE Symp. Comput. Intell. Appl. Smart Grid (CIASG), Apr. 2011,
process regression,’’ Renew. Sustain. Energy Rev., vol. 108, pp. 513–538, pp. 1–8.
Jul. 2019. [138] P. Faria, J. Soares, Z. Vale, H. Morais, and T. Sousa, ‘‘Modified
[116] L. Gigoni, A. Betti, E. Crisostomi, A. Franco, M. Tucci, F. Bizzarri, particle swarm optimization applied to integrated demand response and
and D. Mucci, ‘‘Day-ahead hourly forecasting of power generation DG resources scheduling,’’ IEEE Trans. Smart Grid, vol. 4, no. 1,
from photovoltaic plants,’’ IEEE Trans. Sustain. Energy, vol. 9, no. 2, pp. 606–616, Mar. 2013.
pp. 831–842, Oct. 2018. [139] F. Pereira, J. Soares, P. Faria, and Z. Vale, ‘‘Quantum particle swarm
[117] K. Wang, X. Qi, and H. Liu, ‘‘A comparison of day-ahead photovoltaic optimization applied to distinct remuneration approaches in demand
power forecasting models based on deep learning neural network,’’ Appl. response programs,’’ in Proc. IEEE Symp. Ser. Comput. Intell., Dec. 2015,
Energy, vol. 251, Oct. 2019, Art. no. 113315. pp. 1553–1560.
[118] J. Jeong and H. Kim, ‘‘Multi-site photovoltaic forecasting exploiting [140] M. A. A. Pedrasa, T. D. Spooner, and I. F. MaxGill, ‘‘Scheduling of
space-time convolutional neural network,’’ Energies, vol. 12, no. 23, demand side resources using binary particle swarm optimization,’’ IEEE
p. 4490, Nov. 2019. Trans. Power Syst., vol. 24, no. 3, pp. 1173–1181, Aug. 2009.
[119] A. Tascikaraoglu and M. Uzunoglu, ‘‘A review of combined approaches [141] T. Medved, G. Artač, and A. F. Gubina, ‘‘The use of intelligent
for prediction of short-term wind speed and power,’’ Renew. Sustain. aggregator agents for advanced control of demand response,’’ WIREs
Energy Rev., vol. 34, pp. 243–254, Jun. 2014. Energy Environ., vol. 7, no. 3, May 2018, Art. no. e287.
[120] A. P. Marugán, F. P. G. Márquez, J. M. P. Perez, and D. Ruiz-Hernández, [142] L. A. Hurtado, E. Mocanu, P. H. Nguyen, M. Gibescu, and
‘‘A survey of artificial neural network in wind energy systems,’’ Appl. R. I. G. Kamphuis, ‘‘Enabling cooperative behavior for building
Energy, vol. 228, pp. 1822–1836, Oct. 2018. demand response based on extended joint action learning,’’ IEEE Trans.
[121] H. Z. Wang, G.-Q. Li, G.-B. Wang, J.-C. Peng, H. Jiang, and Y.-T. Liu, Ind. Informat., vol. 14, no. 1, pp. 127–136, Jan. 2018.
‘‘Deep learning based ensemble approach for probabilistic wind power [143] J. Soares, Z. Vale, H. Morais, and N. Borges, ‘‘Demand response in
forecasting,’’ Appl. Energy, vol. 188, pp. 56–70, Feb. 2017. electric vehicles management optimal use of end-user contracts,’’ in Proc.
[122] A. T. Eseye, J. Zhang, and D. Zheng, ‘‘Short-term photovoltaic solar 14th Mex. Int. Conf. Artif. Intell. (MICAI), Oct. 2015, pp. 122–128.
power forecasting using a hybrid wavelet-PSO-SVM model based on [144] T. Shoji, W. Hirohashi, Y. Fujimoto, and Y. Hayashi, ‘‘Home energy man-
SCADA and meteorological information,’’ Renew. Energy, vol. 118, agement based on Bayesian network considering resident convenience,’’
pp. 357–367, Apr. 2018. in Proc. Int. Conf. Probabilistic Methods Appl. Power Syst. (PMAPS),
[123] H. Liu, X.-W. Mi, and Y.-F. Li, ‘‘Wind speed forecasting method based on Jul. 2014, pp. 1–6.
deep learning strategy using empirical wavelet transform, long short term [145] D. O’Neill, M. Levorato, A. Goldsmith, and U. Mitra, ‘‘Residential
memory neural network and Elman neural network,’’ Energy Convers. demand response using reinforcement learning,’’ in Proc. 1st IEEE Int.
Manage., vol. 156, pp. 498–514, Jan. 2018. Conf. Smart Grid Commun., Oct. 2010, pp. 409–414.
[124] H. Shaker, H. Zareipour, and D. Wood, ‘‘Estimating power generation [146] Y. H. Lin and M. S. Tsai, ‘‘An advanced home energy management system
of invisible solar sites using publicly available data,’’ IEEE Trans. Smart facilitated by nonintrusive load monitoring with automated multiobjective
Grid, vol. 7, no. 5, pp. 2456–2465, Sep. 2016. power scheduling,’’ IEEE Trans. Smart Grid, vol. 6, no. 4, pp. 1839–1851,
[125] R. Ahmadiahangar, T. Häring, A. Rosin, T. Korotko, and J. Martins, Jul. 2015.
‘‘Residential load forecasting for flexibility prediction using machine [147] J. M. Veras, I. R. S. Silva, P. R. Pinheiro, R. A. L. Rabêlo, A. F. S. Veloso,
learning-based regression model,’’ in Proc. IEEE Int. Conf. Environ. F. A. S. Borges, and J. J. P. C. Rodrigues, ‘‘A multi-objective demand
Electr. Eng. IEEE Ind. Commercial Power Syst. Eur., Jun. 2019, pp. 1–4. response optimization model for scheduling loads in a home energy
[126] M. Vallés, ‘‘Efficient implementation and potential benefits of demand management system,’’ Sensors, vol. 18, no. 10, p. 3207, 2018.
response in electricity distribution networks,’’ Universidad Pontificia [148] C. Patyn, F. Ruelens, and G. Deconinck, ‘‘Comparing neural architectures
Comillas Madrid, Spain, 2017. for demand response through model-free reinforcement learning for
[127] J. Iria, F. Soares, and M. Matos, ‘‘Optimal bidding strategy for an heat pump control,’’ in Proc. IEEE Int. Energy Conf. (ENERGYCON),
aggregator of prosumers in energy and secondary reserve markets,’’ Appl. Jun. 2018, pp. 1–6.
Energy, vol. 238, pp. 1361–1372, Mar. 2019. [149] S. Bahrami, V. W. S. Wong, and J. Huang, ‘‘An online learning algorithm
[128] S. Ø. Ottesen, A. Tomasgard, and S.-E. Fleten, ‘‘Multi market bidding for demand response in smart grid,’’ IEEE Trans. Smart Grid, vol. 9, no. 5,
strategies for demand side flexibility aggregators in electricity markets,’’ pp. 4712–4725, Sep. 2018.
Energy, vol. 149, pp. 120–134, Apr. 2018. [150] S.-F. Kazemi, S.-A. Motamedi, and S. Sharifian, ‘‘A home energy
[129] A. Lucas, L. Jansen, N. Andreadou, E. Kotsakis, and M. Masera, ‘‘Load management system using gray wolf optimizer in smart grids,’’ in Proc.
flexibility forecast for DR using non-intrusive load monitoring in the 2nd Conf. Swarm Intell. Evol. Comput. (CSIEC), Mar. 2017, pp. 159–164.
residential sector,’’ Energies, vol. 12, no. 14, p. 2725, Jul. 2019. [151] D. L. Cavalca, G. Spavieri, and R. A. S. Fernandes, ‘‘Comparative
[130] S. Dalhues, Y. Zhou, O. Pohl, F. Rewald, F. Erlemeyer, D. Schmid, analysis between particle swarm optimization algorithms applied to price-
J. Zwartscholten, Z. Hagemann, C. Wagner, D. M. Gonzalez, H. Liu, based demand response,’’ in Proc. Int. Conf. Artif. Intell. Soft Comput.,
M. Zhang, J. Liu, C. Rehtanz, Y. Li, and Y. Cao, ‘‘Research and practice 2018, pp. 323–332.
of flexibility in distribution systems: A review,’’ CSEE J. Power Energy [152] V. Margaret and K. U. Rao, ‘‘Demand response for residential loads using
Syst., vol. 5, no. 3, pp. 285–294, Sep. 2019. artificial bee colony algorithm to minimize energy cost,’’ in Proc. IEEE
[131] G. Krishnadas and A. Kiprakis, ‘‘A machine learning pipeline for Region 10 Conf., Nov. 2015, pp. 1–5.
demand response capacity scheduling,’’ Energies, vol. 13, no. 7, p. 1848, [153] S. Sen, S. Chanda, S. Sengupta, and A. De, ‘‘Demand response governed
Apr. 2020. swarm intelligent grid scheduling framework for social welfare,’’ Int. J.
[132] G. Kotsis, I. Moschos, C. Corchero, and M. Cruz-Zambrano, ‘‘Demand Electr. Power Energy Syst., vol. 78, pp. 783–792, Jun. 2016.
aggregator flexibility forecast: Price incentives sensitivity assessment,’’ [154] I. R. S. da Silva, R. D. A. L. Rabêlo, J. J. P. C. Rodrigues, P. Solic,
in Proc. 12th Int. Conf. Eur. Energy Market (EEM), May 2015, pp. 1–5. and A. Carvalho, ‘‘A preference-based demand response mechanism
[133] M. Vellei, J. Le Dréau, and S. Y. Abdelouadoud, ‘‘Predicting the demand for energy management in a microgrid,’’ J. Cleaner Prod., vol. 255,
flexibility of wet appliances at national level: The case of France,’’ Energy May 2020, Art. no. 120034.
Buildings, vol. 214, May 2020, Art. no. 109900. [155] M. V. Goubko, S. O. Kuznetsov, A. A. Neznanov, and D. I. Ignatov,
[134] A. Pirvaram, S. M. Sadrameli, and L. Abdolmaleki, ‘‘Energy management ‘‘Bayesian learning of consumer preferences for residential demand
of a household refrigerator using eutectic environmental friendly PCMs response,’’ IFAC-PapersOnLine, vol. 49, no. 32, pp. 24–29, 2016.
in a cascaded condition,’’ Energy, vol. 181, pp. 321–330, Aug. 2019. [156] Z. Wen, D. O’Neill, and H. Maei, ‘‘Optimal demand response using
[135] C. Ziras, C. Heinrich, M. Pertl, and H. W. Bindner, ‘‘Experimen- device-based reinforcement learning,’’ IEEE Trans. Smart Grid, vol. 6,
tal flexibility identification of aggregated residential thermal loads no. 5, pp. 2312–2324, Sep. 2015.
using behind-the-meter data,’’ Appl. Energy, vol. 242, pp. 1407–1421, [157] F. Ruelens, B. J. Claessens, S. Vandael, B. De Schutter, R. Babuska, and
May 2019. R. Belmans, ‘‘Residential demand response of thermostatically controlled
[136] J.-H. Kim and A. Shcherbakova, ‘‘Common failures of demand loads using batch reinforcement learning,’’ IEEE Trans. Smart Grid,
response,’’ Energy, vol. 36, no. 2, pp. 873–880, Feb. 2011. vol. 8, no. 5, pp. 2149–2159, Sep. 2017.

1502 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

[158] K. Dehghanpour, M. H. Nehrir, J. W. Sheppard, and N. C. Kelly, [180] Y. Xiong, B. Wang, C.-C. Chu, and R. Gadh, ‘‘Vehicle grid integration
‘‘Agent-based modeling of retail electrical energy markets with demand for demand response with mixture user model and decentralized
response,’’ IEEE Trans. Smart Grid, vol. 9, no. 4, pp. 3465–3475, optimization,’’ Appl. Energy, vol. 231, pp. 481–493, Dec. 2018.
Jul. 2018. [181] S. Ikeda and H. Nishi, ‘‘Sparse-coding-based household clustering for
[159] Y. M. Lee, R. Horesh, and L. Liberti, ‘‘Simulation and optimization demand response services,’’ in Proc. IEEE 25th Int. Symp. Ind. Electron.
of energy efficient operation of HVAC system as demand response (ISIE), Jun. 2016, pp. 744–749.
with distributed energy resources,’’ in Proc. Winter Simul. Conf. (WSC), [182] C. Develder, N. Sadeghianpourhamami, M. Strobbe, and N. Refa,
Dec. 2015, pp. 991–999. ‘‘Quantifying flexibility in EV charging as DR potential: Analysis of
[160] T. Jiang, ‘‘Multi-objective optimal scheduling method for regional two real-world data sets,’’ in Proc. IEEE Int. Conf. Smart Grid Commun.
photovoltaic-storage-charging integrated system participating in demand (SmartGridComm), Nov. 2016, pp. 600–605.
response,’’ in Proc. 8th Renew. Power Gener. Conf. (RPG), 2019, pp. 1–8. [183] V. Trovato, S. H. Tindemans, and G. Strbac, ‘‘Leaky storage model
[161] I. Dusparic, C. Harris, A. Marinescu, V. Cahill, and S. Clarke, ‘‘Multi- for optimal multi-service allocation of thermostatic loads,’’ IET Gener.,
agent residential demand response based on load forecasting,’’ in Proc. Transmiss. Distrib., vol. 10, no. 3, pp. 585–593, Feb. 2016.
1st IEEE Conf. Technol. Sustainability (SusTech), Aug. 2013, pp. 90–96. [184] M. Alizadeh, A. Scaglione, A. Goldsmith, and G. Kesidis, ‘‘Capturing
[162] O. Hafez and K. Bhattacharya, ‘‘Integrating EV charging stations as smart aggregate flexibility in demand response,’’ in Proc. 53rd IEEE Conf.
loads for demand response provisions in distribution systems,’’ IEEE Decis. Control, Dec. 2014, pp. 6439–6445.
Trans. Smart Grid, vol. 9, no. 2, pp. 1096–1106, Mar. 2018. [185] J. Iria and F. Soares, ‘‘A cluster-based optimization approach to support
[163] M. Waseem, Z. Lin, S. Liu, Z. Zhang, T. Aziz, and D. Khan, ‘‘Fuzzy the participation of an aggregator of a larger number of prosumers
compromised solution-based novel home appliances scheduling and in the day-ahead energy market,’’ Electr. Power Syst. Res., vol. 168,
demand response with optimal dispatch of distributed energy resources,’’ pp. 324–335, Mar. 2019.
Appl. Energy, vol. 290, May 2021, Art. no. 116761. [186] Y. Gao, B. Foggo, and N. Yu, ‘‘A physically inspired data-driven model
[164] L. Jia, Q. Zhao, and L. Tong, ‘‘Retail pricing for stochastic demand for electricity theft detection with smart meter data,’’ IEEE Trans. Ind.
with unknown parameters: An online machine learning approach,’’ in Informat., vol. 15, no. 9, pp. 5076–5088, Sep. 2019.
Proc. 51st Annu. Allerton Conf. Commun., Control, Comput. (Allerton), [187] R. Razavi, A. Gharipour, M. Fleury, and I. J. Akpan, ‘‘A practical feature-
Oct. 2013, pp. 1353–1358. engineering framework for electricity theft detection in smart grids,’’
[165] M. J. Alves, C. H. Antunes, and P. Carrasqueira, ‘‘A hybrid genetic Appl. Energy, vol. 238, pp. 481–494, Mar. 2019.
algorithm for the interaction of electricity retailers with demand [188] J. Y. Kim, Y. M. Hwang, Y. G. Sun, I. Sim, D. I. Kim, and X. Wang,
response,’’ in Proc. Eur. Conf. Appl. Evol. Comput., 2016, pp. 459–474. ‘‘Detection for non-technical loss by smart energy theft with intermediate
[166] F. Meng, X.-J. Zeng, Y. Zhang, C. J. Dent, and D. Gong, ‘‘An monitor meter in smart grid,’’ IEEE Access, vol. 7, pp. 129043–129053,
integrated optimization + learning approach to optimal dynamic pricing 2019.
for the retailer with multi-type customers in smart grids,’’ Inf. Sci., [189] Z. Chen, D. Meng, Y. Zhang, T. Xin, and D. Xiao, ‘‘Electricity theft
vols. 448–449, pp. 215–232, Jun. 2018. detection using deep bidirectional recurrent neural network,’’ in Proc.
[167] F. Lopes, C. Ilco, and J. Sousa, ‘‘Bilateral negotiation in energy markets: 22nd Int. Conf. Adv. Commun. Technol. (ICACT), Feb. 2020, pp. 401–406.
Strategies for promoting demand response,’’ in Proc. 10th Int. Conf. Eur.
[190] P. P. Biswas, H. Cai, B. Zhou, B. Chen, D. Mashima, and V. W. Zheng,
Energy Market (EEM), May 2013, pp. 1–6.
‘‘Electricity theft pinpointing through correlation analysis of master and
[168] S. Lin, F. Li, E. Tian, Y. Fu, and D. Li, ‘‘Clustering load profiles for
individual meter readings,’’ IEEE Trans. Smart Grid, vol. 11, no. 4,
demand response applications,’’ IEEE Trans. Smart Grid, vol. 10, no. 2,
pp. 3031–3042, Jul. 2020.
pp. 1599–1607, Mar. 2019.
[191] D. Yao, M. Wen, X. Liang, Z. Fu, K. Zhang, and B. Yang, ‘‘Energy
[169] H.-A. Cao, C. Beckel, and T. Staake, ‘‘Are domestic load profiles stable
theft detection with energy privacy preservation in the smart grid,’’ IEEE
over time? An attempt to identify target households for demand side
Internet Things J., vol. 6, no. 5, pp. 7659–7669, Oct. 2019.
management campaigns,’’ in Proc. 39th Annu. Conf. IEEE Ind. Electron.
[192] Y. Liu, T. Liu, H. Sun, K. Zhang, and P. Liu, ‘‘Hidden electricity theft
Soc., Nov. 2013, pp. 4733–4738.
by exploiting multiple-pricing scheme in smart grids,’’ IEEE Trans. Inf.
[170] D. Koolen, N. Sadat-Razavi, and W. Ketter, ‘‘Machine learning for
Forensics Security, vol. 15, pp. 2453–2468, 2020.
identifying demand patterns of home energy management systems with
dynamic electricity pricing,’’ Appl. Sci., vol. 7, no. 11, p. 1160, 2017. [193] D. Hock, M. Kappes, and B. Ghita, ‘‘Using multiple data sources to detect
[171] J. Kwac, J. Flora, and R. Rajagopal, ‘‘Household energy consumption manipulated electricity meter by an entropy-inspired metric,’’ Sustain.
segmentation using hourly data,’’ IEEE Trans. Smart Grid, vol. 5, no. 1, Energy, Grids Netw., vol. 21, Mar. 2020, Art. no. 100290.
pp. 420–430, Jan. 2014. [194] A. Kumari, R. Gupta, S. Tanwar, and N. Kumar, ‘‘Blockchain and AI
[172] S. Haben, C. Singleton, and P. Grindrod, ‘‘Analysis and clustering of amalgamation for energy cloud management: Challenges, solutions, and
residential customers energy behavioral demand using smart meter data,’’ future directions,’’ J. Parallel Distrib. Comput., vol. 143, pp. 148–166,
IEEE Trans. Smart Grid, vol. 7, no. 1, pp. 136–144, Jan. 2016. Sep. 2020.
[173] A. C. Varghese, V. Padmini, G. Kumar, and S. A. Khaparde, ‘‘Smart [195] Z. Khalid, G. Abbas, M. Awais, T. Alquthami, and M. B. Rasheed,
grid consumer behavioral model using machine learning,’’ in Proc. IEEE ‘‘A novel load scheduling mechanism using artificial neural network
Innov. Smart Grid Technol. Asia (ISGT Asia), May 2018, pp. 734–739. based customer profiles in smart grid,’’ Energies, vol. 13, no. 5, p. 1062,
[174] S. Ahmed and F. Bouffard, ‘‘Building load management clusters using Feb. 2020.
reinforcement learning,’’ in Proc. 8th IEEE Annu. Inf. Technol., Electron. [196] B. Yuce, Y. Rezgui, and M. Mourshed, ‘‘ANN–GA smart appliance
Mobile Commun. Conf. (IEMCON), Oct. 2017, pp. 372–377. scheduling for optimised energy management in the domestic sector,’’
[175] T. Chen, K. Qian, A. Mutanen, B. Schuller, P. Järventausta, and W. Su, Energy Buildings, vol. 111, pp. 311–325, Jan. 2016.
‘‘Classification of electricity customer groups towards individualized [197] F. Smarra, A. Jain, T. de Rubeis, D. Ambrosini, A. D’Innocenzo, and
price scheme design,’’ in Proc. North Amer. Power Symp. (NAPS), R. Mangharam, ‘‘Data-driven model predictive control using random
Sep. 2017, pp. 1–4. forests for building energy optimization and climate control,’’ Appl.
[176] J. Spínola, P. Faria, and Z. Vale, ‘‘Energy resource aggregator managing Energy, vol. 226, pp. 1252–1272, Sep. 2018.
active consumer demand programs,’’ in Proc. IEEE Symp. Ser. Comput. [198] M. S. Ahmed, A. Mohamed, R. Z. Homod, and H. Shareef, ‘‘Hybrid
Intell. (SSCI), Dec. 2016, pp. 1–6. LSA-ANN based home energy management scheduling controller for
[177] J. Spínola, R. Faia, P. Faria, and Z. Vale, ‘‘Clustering optimization of residential demand response strategy,’’ Energies, vol. 9, no. 9, p. 716,
distributed energy resources in support of an aggregator,’’ in Proc. IEEE 2016.
Symp. Ser. Comput. Intell. (SSCI), Nov. 2017, pp. 1–6. [199] S. Bourhnane, M. R. Abid, R. Lghoul, K. Zine-Dine, N. Elkamoun, and
[178] D. Çelik, M. E. Meral, and M. Waseem, ‘‘Investigation and analysis of D. Benhaddou, ‘‘Machine learning for energy consumption prediction
effective approaches, opportunities, bottlenecks and future potential capa- and scheduling in smart buildings,’’ Social Netw. Appl. Sci., vol. 2, no. 2,
bilities for digitalization of energy systems and sustainable development pp. 1–10, Feb. 2020.
goals,’’ Electric Power Syst. Res., vol. 211, Oct. 2022, Art. no. 108251. [200] E. Koliou, C. Eid, J. P. Chaves-Ávila, and R. A. Hakvoort, ‘‘Demand
[179] X. Lu, K. Li, H. Xu, F. Wang, Z. Zhou, and Y. Zhang, ‘‘Fundamentals and response in liberalized electricity markets: Analysis of aggregated load
business model for resource aggregator of demand response in electricity participation in the German balancing mechanism,’’ Energy, vol. 71,
markets,’’ Energy, vol. 204, Aug. 2020, Art. no. 117885. pp. 245–254, Jul. 2014.

VOLUME 11, 2023 1503

 M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

[201] L. C. Casals, M. Barbero, and C. Corchero, ‘‘Reused second life batteries [222] A. Silis, V. Lavrinovics, G. Junghans, and A. Sauhats, ‘‘Benefits of
for aggregated demand response services,’’ J. Cleaner Prod., vol. 212, electricity industry switching from fixed to spot-linked end-user prices,’’
pp. 99–108, Mar. 2019. in Proc. 15th Int. Conf. Eur. Energy Market (EEM), Jun. 2018, pp. 1–5.
[202] K. Wohlfarth, E. Worrell, and W. Eichhammer, ‘‘Energy efficiency and [223] L. Sadovica, G. Junghans, A. Sauhats, Z. Broka, K. Baltputnis, and
demand response—Two sides of the same coin?’’ Energy Policy, vol. 137, V. Lavrinovics, ‘‘Case study—Assessing economic potential for demand
Feb. 2020, Art. no. 111070. response in Baltic balancing market,’’ in Proc. IEEE 59th Int. Sci. Conf.
[203] W. Li, P. Xu, X. Lu, H. Wang, and Z. Pang, ‘‘Electricity demand response Power Electr. Eng. Riga Tech. Univ. (RTUCON), Nov. 2018, pp. 1–5.
in China: Status, feasible market schemes and pilots,’’ Energy, vol. 114, [224] L. Sadovica, V. Lavrinovics, A.-S. Sauhats, G. Junghans, and
pp. 981–994, Nov. 2016. K. M. Lehtmets, ‘‘Estimating energy reduction amount in the event
[204] J. Torriti, M. G. Hassan, and M. Leach, ‘‘Demand response experience of demand response activation: Baseline model comparison for the Baltic
in Europe: Policies, programmes and implementation,’’ Energy, vol. 35, states,’’ in Proc. 15th Int. Conf. Eur. Energy Market (EEM), Jun. 2018,
no. 4, pp. 1575–1583, Apr. 2010. pp. 1–5.
[205] I. M. Chatzigeorgiou, D. Manolas, T. Gkaragkouni, and G. T. Andreou, [225] A. Shivakumar, S. Pye, J. Anjo, M. Miller, P. B. Rouelle, M. Densing, and
‘‘Demand response in Greece: An introductory mobile application,’’ in T. Kober, ‘‘Smart energy solutions in the EU: State of play and measuring
Proc. IEEE Int. Conf. Environ. Electr. Eng. IEEE Ind. Commercial Power progress,’’ Energy Strategy Rev., vol. 20, pp. 133–149, Apr. 2018.
Syst. Eur., Jun. 2018, pp. 1–5. [226] Y. Shahzad, H. Javed, H. Farman, J. Ahmad, B. Jan, and M. Zubair,
‘‘Internet of energy: Opportunities, applications, architectures and
[206] D. Pattabiraman, R. H. Lasseter., and T. M. Jahns, ‘‘Comparison of grid
challenges in smart industries,’’ Comput. Electr. Eng., vol. 86, Sep. 2020,
following and grid forming control for a high inverter penetration power
Art. no. 106739.
system,’’ in Proc. IEEE Power Energy Soc. Gen. Meeting (PESGM),
[227] P. K. Khatua, V. K. Ramachandaramurthy, P. Kasinathan, J. Y. Yong,
Aug. 2018, pp. 1–5.
J. Pasupuleti, and A. Rajagopalan, ‘‘Application and assessment of
[207] D. D. Micu, B. Bârgăuan, A. Ceclan, D. Stet, L. Czumbil, A. Catinean,
Internet of Things toward the sustainability of energy systems: Challenges
and A. Ploycarpou, ‘‘On a demand response pilot demonstration in the
and issues,’’ Sustain. Cities Soc., vol. 53, Feb. 2020, Art. no. 101957.
technical university of Cluj–Napoca,’’ in Proc. Int. Conf. Expo. Electr. [228] H. Ma, Y. Zhang, and M. Shen, ‘‘Application and prospect of
Power Eng. (EPE), Oct. 2016, pp. 785–791. supercapacitors in internet of energy (IOE),’’ J. Energy Storage, vol. 44,
[208] P. Warren, ‘‘A review of demand-side management policy in the U.K.,’’ Dec. 2021, Art. no. 103299.
Renew. Sustain. Energy Rev., vol. 29, pp. 941–951, Jan. 2014. [229] F. Wang, X. Lu, X. Chang, X. Cao, S. Yan, K. Li, N. Duić,
[209] M. Goulden, A. Spence, J. Wardman, and C. Leygue, ‘‘Differentiating M. Shafie-Khah, and J. P. S. Catalão, ‘‘Household profile identification
‘the user’ in DSR: Developing demand side response in advanced for behavioral demand response: A semi-supervised learning approach
economies,’’ Energy Policy, vol. 122, pp. 176–185, Nov. 2018. using smart meter data,’’ Energy, vol. 238, Jan. 2022, Art. no. 121728.
[210] C. A. Cardoso, J. Torriti, and M. Lorincz, ‘‘Making demand side response [230] C. Zhao, H. Shi, R. Li, and F. Li, ‘‘Demand side response performance
happen: A review of barriers in commercial and public organisations,’’ assessment: An impact analysis of load profile accuracy on DSR
Energy Res. Social Sci., vol. 64, Jun. 2020, Art. no. 101443. performances,’’ in Proc. IEEE Power Energy Soc. Gen. Meeting,
[211] C. Kuzemko, C. Mitchell, M. Lockwood, and R. Hoggett, ‘‘Policies, Jul. 2015, pp. 1–5.
politics and demand side innovations: The untold story of Germany’s [231] M. Radenković, Z. Bogdanović, M. Despotović-Zrakić, A. Labus, and
energy transition,’’ Energy Res. Social Sci., vol. 28, pp. 58–67, Jun. 2017. S. Lazarević, ‘‘Assessing consumer readiness for participation in IoT-
[212] F. Carus, ‘‘What next for demand side response?’’ Renew. Energy Focus, based demand response business models,’’ Technol. Forecasting Social
vol. 17, no. 1, pp. 28–30, 2016. Change, vol. 150, Jan. 2020, Art. no. 119715.
[213] S. Talari, D. Mende, D. S. Stock, M. Shafie-Khah, and J. P. S. Catalao, [232] G. J. Osório, M. Shafie-Khah, N. G. S. Soares, and J. P. S. Catalão,
‘‘Stochastic demand side management in European zonal price market,’’ ‘‘Optimal dynamic tariffs for flexible ramp market in the presence of
in Proc. Int. Conf. Smart Energy Syst. Technol. (SEST), Sep. 2019, wind power generation and demand response,’’ in Proc. IEEE Int. Conf.
pp. 1–6. Environ. Electr. Eng. IEEE Ind. Commercial Power Syst. Eur., Jun. 2018,
[214] A. K. Mishra, J. Jokisalo, R. Kosonen, T. Kinnunen, M. Ekkerhaugen, pp. 1–5.
H. Ihasalo, and K. Martin, ‘‘Demand response events in district heating: [233] C. Patsonakis, S. Terzi, I. Moschos, D. Ioannidis, K. Votis, and
Results from field tests in a university building,’’ Sustain. Cities Soc., D. Tzovaras, ‘‘Permissioned blockchains and virtual nodes for reinforc-
vol. 47, May 2019, Art. no. 101481. ing trust between aggregators and prosumers in energy demand response
[215] S. Annala, J. Lukkarinen, E. Primmer, S. Honkapuro, K. Ollikka, scenarios,’’ in Proc. IEEE Int. Conf. Environ. Electr. Eng. IEEE Ind.
K. Sunila, and T. Ahonen, ‘‘Regulation as an enabler of demand response Commercial Power Syst. Eur., Jun. 2019, pp. 1–6.
in electricity markets and power systems,’’ J. Cleaner Prod., vol. 195, [234] J. S. Vardakas, N. Zorba, and C. V. Verikoukis, ‘‘A survey on demand
pp. 1139–1148, Sep. 2018. response programs in smart grids: Pricing methods and optimization
[216] S. Annala, G. Mendes, S. Honkapuro, L. Matos, and L. P. Klein, algorithms,’’ IEEE Commun. Surveys Tuts., vol. 17, no. 1, pp. 152–178,
‘‘Comparison of opportunities and challenges in demand response pilots 1st Quart., 2014.
[235] A. Mellit and S. Kalogirou, ‘‘Artificial intelligence and Internet of Things
in Finland and Portugal,’’ in Proc. 15th Int. Conf. Eur. Energy Market
to improve efficacy of diagnosis and remote sensing of solar photovoltaic
(EEM), Jun. 2018, pp. 1–5.
systems: Challenges, recommendations and future directions,’’ Renew.
[217] L. Söder, P. D. Lund, H. Koduvere, T. F. Bolkesjø, G. H. Rossebø,
Sustain. Energy Rev., vol. 143, Jun. 2021, Art. no. 110889.
E. Rosenlund-Soysal, K. Skytte, J. Katz, and D. Blumberga, ‘‘A review
[236] M. Waseem, Z. Lin, S. Liu, I. A. Sajjad, and T. Aziz, ‘‘Optimal GWCSO-
of demand side flexibility potential in northern Europe,’’ Renew. Sustain.
based home appliances scheduling for demand response considering
Energy Rev., vol. 91, pp. 654–664, Aug. 2018.
end-users comfort,’’ Electr. Power Syst. Res., vol. 187, Oct. 2020,
[218] M. Barbero, C. Corchero, L. C. Casals, L. Igualada, and F.-J. Heredia, Art. no. 106477.
‘‘Critical evaluation of European balancing markets to enable the [237] A. Goudarzi, F. Ghayoor, M. Waseem, S. Fahad, and I. Traore, ‘‘A sur-
participation of demand aggregators,’’ Appl. Energy, vol. 264, Apr. 2020, vey on IoT-enabled smart grids: Emerging, applications, challenges,
Art. no. 114707. and outlook,’’ Energies, vol. 15, no. 19, p. 6984, Sep. 2022, doi:
[219] N. Boogen, S. Datta, and M. Filippini, ‘‘Demand-side management 10.3390/en15196984.
by electric utilities in Switzerland: Analyzing its impact on residential [238] J. M. Paredes-Parra, A. J. García-Sánchez, A. Mateo-Aroca, and
electricity demand,’’ Energy Econ., vol. 64, pp. 402–414, May 2017. J. E. Molina-García, ‘‘An alternative Internet-of-Things solution based on
[220] A. Srivastava, S. Van Passel, R. Kessels, P. Valkering, and E. Laes, LoRa for PV power plants: Data monitoring and management,’’ Energies,
‘‘Reducing winter peaks in electricity consumption: A choice experiment vol. 12, no. 5, p. 881, 2019.
to structure demand response programs,’’ Energy Policy, vol. 137, [239] G. V. Bhau, R. G. Deshmukh, T. R. Kumar, S. Chowdhury, Y. Sesharao,
Feb. 2020, Art. no. 111183. and Y. Abilmazhinov, ‘‘IoT based solar energy monitoring system,’’
[221] L. Sadoviča, K. Marcina, V. Lavrinovičs, and G. Junghans, ‘‘Facilitating Mater. Today, Proc., Jul. 2021.
energy system flexibility by demand response in the baltics—Choice of [240] R. Liang, Y. Guo, L. Zhao, and Y. Gao, ‘‘Real-time monitoring
the market model,’’ in Proc. IEEE 58th Int. Sci. Conf. Power Electr. Eng. implementation of PV/T façade system based on IoT,’’ J. Building Eng.,
Riga Tech. Univ. (RTUCON), Oct. 2017, pp. 1–6. vol. 41, Sep. 2021, Art. no. 102451.

1504 VOLUME 11, 2023

M. A. Khan et al.: AI Enabled DR: Prospects and Challenges in Smart Grid Environment

[241] D. D. P. Rani, D. Suresh, P. R. Kapula, C. H. M. Akram, N. Hemalatha, AHMED MOHAMMED SALEH was born in
and P. K. Soni, ‘‘IoT based smart solar energy monitoring systems,’’ Yemen. He received the bachelor’s degree (Hons.)
Mater. Today, Proc., Jul. 2021. in electrical engineering from the University of
[242] C. Tang, M. Liu, Q. Liu, and P. Dong, ‘‘A per-node granularity Aden, Yemen, in 2015. He is currently pursuing
decentralized optimal power flow for radial distribution networks with the Ph.D. degree. He worked as a Laboratory
PV and EV integration,’’ Int. J. Electr. Power Energy Syst., vol. 116, Engineer with the University of Aden, until 2019.
Mar. 2020, Art. no. 105513.
His research interests include smart grid, the
[243] V. Kumar and M. Singh, ‘‘Reactive power compensation using derated
power generation mode of modified P&O algorithm in grid-interfaced hybridization of renewable energy sources, and
PV system,’’ Renew. Energy, vol. 178, pp. 108–117, Nov. 2021. optimization techniques.
[244] E. Kapassa and M. Themistocleous, ‘‘Blockchain technology applied
in IoV demand response management: A systematic literature review,’’
Future Internet, vol. 14, no. 5, p. 136, Apr. 2022.
[245] M. A. Khan, I. A. Sajjad, M. Tahir, and A. Haseeb, ‘‘IoT application for
energy management in smart homes,’’ Eng. Proc., vol. 20, no. 1, p. 43,
2022, doi: 10.3390/engproc2022020043.
[246] S. Zhu, M. Song, M. K. Lim, J. Wang, and J. Zhao, ‘‘The development of
energy blockchain and its implications for China’s energy sector,’’ Resour.
Policy, vol. 66, Jun. 2020, Art. no. 101595.
[247] Z. M. Haider, K. K. Mehmood, S. U. Khan, M. O. Khan, A. Wadood,
and S.-B. Rhee, ‘‘Optimal management of a distribution feeder during MUHAMMAD WASEEM received the B.Sc. and
contingency and overload conditions by harnessing the flexibility of smart M.Sc. degrees in electrical engineering from the
loads,’’ IEEE Access, vol. 9, pp. 40124–40139, 2021. University of Engineering and Technology Taxila,
[248] M. S. Javadi, A. E. Nezhad, P. H. J. Nardelli, M. Gough, M. Lotfi, Pakistan, in 2012 and 2017, respectively, and
S. Santos, and J. P. S. Catalão, ‘‘Self-scheduling model for home energy the Ph.D. degree in electrical engineering from
management systems considering the end-users discomfort index within
Zhejiang University, Hangzhou, China, in 2022.
price-based demand response programs,’’ Sustain. Cities Soc., vol. 68,
He is currently with the Department of Electri-
May 2021, Art. no. 102792.
[249] P. Pinson, H. Madsen, H. A. Nielsen, G. Papaefthymiou, and B. Klöckl, cal Engineering, University of Engineering and
‘‘From probabilistic forecasts to statistical scenarios of short-term wind Technology Taxila. His research interests include
power production,’’ Wind Energy, vol. 12, no. 1, pp. 51–62, 2009. power system analysis, demand side management,
[250] Y. Simmhan, S. Aman, A. Kumbhare, R. Liu, S. Stevens, Q. Zhou, and and smart grid. He is a Lifetime Member of the Pakistan Engineering Council
V. Prasanna, ‘‘Cloud-based software platform for big data analytics in (PEC). He has acted as a referee for various international conferences and
smart grids,’’ Comput. Sci. Eng., vol. 15, no. 4, pp. 38–47, Jul. 2013. journals.
[251] E. Mocanu, P. H. Nguyen, W. L. Kling, and M. Gibescu, ‘‘Unsupervised
energy prediction in a smart grid context using reinforcement cross-
building transfer learning,’’ Energy Buildings, vol. 116, pp. 646–655,
Mar. 2016.

MUHAMMAD ADNAN KHAN received the INTISAR ALI SAJJAD (Member, IEEE) received
B.Sc. and M.Sc. degrees in electrical engineering the Ph.D. degree in electrical engineering from
from the University of Engineering and Technol- the Politecnico di Torino, Turin, Italy, in 2015.
ogy Taxila, Pakistan, in 2019 and 2022, respec- He is currently working as an Associate Professor
tively. He is currently with the Department of with the Department of Electrical Engineering,
Electrical Engineering, University of Engineering University of Engineering and Technology Taxila,
and Technology Taxila. His research interests Pakistan. His current research interests include
include renewable energy, smart grids, distributed smart buildings, AI applications in power systems,
energy resources integration, integration of renew- and load management. He is a Lifetime Member of
able energy in smart grids, and demand side the Pakistan Engineering Council (PEC). He has
management. He is a Lifetime Member of the Pakistan Engineering acted as a referee for various international conferences and journals.
Council (PEC).

VOLUME 11, 2023 1505