sustainability

Review
Review of Machine Learning Techniques for Power Quality
Performance Evaluation in Grid-Connected Systems
Ramya Kuppusamy 1 , Srete Nikolovski 2 and Yuvaraja Teekaraman 3, *

1 Department of Electrical and Electronics Engineering, Sri Sairam College of Engineering,

Bangalore 562 106, India; [email protected]
2 EPIK d.o.o. Nasice, 31500 Našice, Croatia; [email protected]
3 School of Engineering and Computing, American International University (AIU), Al Jahra 003200, Kuwait
* Correspondence: [email protected]

Abstract: In the current energy usage scenario, the demands on energy load and the tariffs on
the usage of electricity are two main areas that require a lot of attention. Energy forecasting is
an ideal solution that would help us to better understand future needs and formulate solutions
accordingly. Some important factors to investigate are the quantity and quality of smart grids as they
are significantly influenced by the transportation, storage, and load management of energy. This
research work is a review of various machine learning algorithms for energy grid applications like
energy consumption, production, energy management, design, vehicle-to-grid transfers, and demand
response. Ranking is performed with the help of key parameters and is evaluated using the Rapid
Miner tool. The proposed manuscript uses various machine learning techniques for the evaluation of
power quality performance to validate an efficient algorithm ranking in a grid-connected system for
energy management applications. The use of renewable energy resources in grid-connected systems is
more common in modern power systems. Universally, the energy usage sector (commercial and non-
commercial) is undergoing an increase in demand for energy utilization that has substantial economic
and ecological consequences. To overcome these issues, an integrated, ecofriendly, and smart system
that meets the high energy demands is implemented in various buildings and other grid-connected
Citation: Kuppusamy, R.; Nikolovski, applications. Among various machine learning techniques, an evaluation of seven algorithms—Naïve
S.; Teekaraman, Y. Review of Bayes, artificial neural networks, linear regression, support vector machine, Q-learning, Gaussian
Machine Learning Techniques for mixture model, and principle component analysis—was conducted to determine which algorithm is
Power Quality Performance the most effective in predicting energy balance. Among these algorithms, the decision tree, linear
Evaluation in Grid-Connected regression, and neural networks had more accurate results than the other algorithms used. As a result
Systems. Sustainability 2023, 15, 15055. of this research, a proposal for energy forecast, energy balance, and management was compiled. A
https://doi.org/10.3390/
comparative statement of various algorithms concludes with results which suit energy management
su152015055
applications with high accuracy and low error rates.
Academic Editors: Nicola Delmonte,
Andrea Toscani and Paolo Cova Keywords: machine learning techniques; power quality; energy management

Received: 7 September 2023

Revised: 4 October 2023
Accepted: 13 October 2023
Published: 19 October 2023
1. Introduction
In the current context, it is important to conserve and optimally use energy as it
is wasted in large amounts every day due to the increased demand for energy and an
inability to efficiently use it. Thus arises a need for the efficient management of energy.
Copyright: © 2023 by the authors.
Energy can be generated in various ways and purchased from national grids. Usage of
Licensee MDPI, Basel, Switzerland.
energy varies from day to day, depending on the requirements and weather. In order
This article is an open access article
to acquire a clear picture of day-to-day energy requirements, energy usage needs to be
distributed under the terms and
monitored and analyzed. This analysis could be a tedious process due to the varying
conditions of the Creative Commons
patterns of energy usage and the large amounts of data that need to be analyzed. In this
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
review, several journals and papers are grouped according to their respective energy grid
4.0/).
applications, namely, consumption, production, demand response, energy management,

Sustainability 2023, 15, 15055. https://doi.org/10.3390/su152015055 https://www.mdpi.com/journal/sustainability

Sustainability 2023, 15, 15055 2 of 29

design, and vehicle-to-grid transfer. Energy consumption in buildings is an important

energy grid application and is reviewed as follows. The paper from Bagnasco et al., 2015 [1],
gives us the forecasting results of energy consumption for a hospital facility and discusses
the use case in a hospital with reference to energy forecasting and energy consumption.
Abdullatif E., 2004 [2], discusses energy conversion and management for the usage of load
in buildings. An artificial intelligence approach can be used to analyze algorithms and come
to conclusions since it can deal with large datasets efficiently [3]. Alberto Hernandez Neto
et al., 2008 [4], discussed a comparative study to anticipate the energy consumption in an
administration building using an ANN model with a thorough building HVAC design and
simulation software. In the paper by Alejandro et al., 2016 [5], an integrated architecture
is proposed that uses the existing available data and unstructured information for better
energy consumption prediction. Betul et al., 2009 [6], predicted a building’s energy forecast
using three-layered backpropagation artificial neural networks. Joaquim Massana et al.,
2016 [7], proposed a real case study in Girona’s university by applying the forecasting
method to non-residential buildings to realize the performance discrimination between
the artificial occupancy attributes. Carlos Roldán-Blay et al., 2013 [8], explored various
techniques in artificial neural networks to predict electrical power load in buildings to
forecast the time–temperature curve, energy usage, and performance data. N. Scarlat et al.,
2014 [9], constructed an ANN model to predict the existing building heat load and its
energy consumption; the obtained results were compared with an energy simulation tool
called KEP-IYTE-ESS. D. Hawkins et al., 2012 [10], analyzed the distribution of energy
usage in university buildings focusing on internal building environment and activity. The
paper from Dandan Liu et al., 2013 [11], developed a support vector regression approach
with a radial function; this was applied to predict the amount of energy consumption
and the obtained results were compared using non-linear data regression. The developed
prediction model used support vector regression with a radial function to forecast the
energy required and assess the energy management plan. Although there are various
options for managing energy, the need for more accurate prediction prevails in order to
handle energy requirements and distribution. A loss in revenue results from unused energy
and this can be avoided by predicting the amount of energy required and purchasing
only the required amount of energy. In the given scenario, solar panels, cogeneration,
consumption from buildings, and the national grid are connected to the smart grid. Through
accurate prediction of unused energy, the wastage of energy can be controlled, and loss
of revenue can also be prevented at a high level. Energy consumption prediction leads to
many options in choosing a machine learning algorithm and produces a data analysis tool
that can be used to address the following concerns.
1. Reviewing various machine learning algorithms with some key parameters for each
energy grid application, which is essential in finding the ideal algorithm;
2. Identification of patterns in the energy consumption of buildings in a given scenario;
3. Correlation analysis between the various features in the scenario;
4. Usage of the best algorithm is decided according to the consumption prediction results
for energy management applications.
Various machine learning algorithms are used for prediction analysis of energy grid
applications. The major work carried out in this manuscript is a review analysis of various
possible machine learning algorithms to filter them and obtain the ideal algorithm for
implementation in the respective energy grid applications. The requirement in any given
technical framework is the implementation of better energy management programs where
automation and user discomfort minimization are needed. Machine learning algorithms
provide a great contribution to energy management programs through prediction analysis.
David Solomon et al. [12] forecast the peak demand in a larger building to coun-
terbalance the smart grid scenario. In the paper by Li Xuemei et al., 2010 [13], a load
forecasting approach and its experimental result are presented; they used fuzzy support
vector machines and a fuzzy C-mean clustering algorithm to indicate that the technique
can be used as an effective approach to achieve short-term cooling. Deyin Zhao et al.,
Sustainability 2023, 15, 15055 3 of 29

2016 [14], proposed a black box model using multivariant regression in VRV systems in
office buildings to predict the total energy consumption. Faran et al., 2018 [15], presents a
summary of resource management in a cellular base station powered by renewable sources
and a thorough analysis of power usage and optimization is carried out to reduce costs
and greenhouse gas emissions. This contributes to green communication towards climate
change. Guang Shi et al., 2016 [16], and Guillermo Escrivá et al., 2011 [17], discussed an
ANN method for the short-term prediction of the consumption of total power in buildings
with several independent processes, while considering the load and end users. H. X. Zhao
et al., 2010 [18], propose an SVM model to forecast power usage in multiple buildings.
Kangji Li et al., 2011 [19], forecast the load demand in buildings using a hybrid GA-ANFIS
model. Ivan Korolija et al., 2013 [20,21], described the construction of an ANN and regres-
sion model to predict the annual energy required for heating and cooling in office buildings.
Jin Yang et al., 2005 [22], described static ANN models to forecast energy consumption with
all the independent parameters known at time t. In the paper by Tiberia et al., 2013 [23],
an approach to determining heating energy demand for fast prediction using multiple
regression model is presented.
Regression models also yield good results for forecasting energy demand in buildings;
this study demonstrates the usage of a multiple regression model to achieve good accuracy.
In the paper by V.A. Kamadev et al., 2010 [24], forecast of energy consumption in shopping
malls is perceived using connectionist systems. In the paper by Yixuan et al., 2018 [25],
data-driven statistics are demonstrated different buildings for energy-related applications,
such as energy prediction, energy consumption, load forecasting, energy pattern profiling,
benchmarking for building stocks, and guidelines drafting for global retrofit strategies. I the
paper by Yoseba et al., 2011 [26], air-conditioned non-residential and commercial buildings
for short-term load forecasting are presented; it is demonstrated and observed that short-
term forecasting yields better results than long-term forecasting in the future. Young et al.,
2016 [27], proposed adaptive training methods and a data-driven forecasting model for
determining the day-ahead electricity usage of buildings at to a 15-minute precision level
to predict electricity consumption. Yumiko Iwafune et al., 2014 [28], examined different
forecasting methods to support energy management in a house on a day-ahead basis.
Zilong et al., 2017 [29], collected a dataset from hospital buildings and analyzed the data to
forecast the energy consumption in the building. Analyses and surveys are performed in
hospital buildings to determine their energy consumption applications. Adriana Chis et al.,
2016 [30], with a previous tariff as reference, simulated and observed various charging
scenarios for day-to-day historical time frames; they determined that the long-term costs
of individual plug-in electric vehicles were reduced. Guang Shi et al., 2017 [31] describe
the implementation of an echo state network technique that uses the Q function to control
and determine the status of battery charging and discharging in offices using renewable
energy. Frederik et al., 2016 [32], present a Q-learning approach along with an online
Markov chain; in the study, this is used to evaluate and display the best methodologies for
tracking hybrid electric vehicles. José R. Vázquez Canteli et al., [33], present a brief review
of various machine learning algorithms and modelling techniques. Deep reinforcement
learning is a version of deep learning that is used for detailed analysis before obtaining
results. An efficient energy management approach for hybrid electric vehicles is obtained
using deep reinforcement learning. Nora El-Nohari et al. [34] examined the impact of
demand response and load shifting in residential houses. The derived simulation results
provided an overview of data analytics and a prediction of the energy consumption in
buildings. Xiaoshun et al., 2017 [35], present a use of the Stackelberg game, which is
utilized to evaluate the supply–demand in a smart grid that operates through deep transfer
Q-learning. Rui Xiong et al., 2018 [36] demonstrate a power management methodology
through a validation of a real-time dataset using a battery and an ultracapacitor in-the-loop
approach. The designed model for any system must be ecofriendly and reliable. A survey of
the most significant energy grid applications and their designs is presented in the following
works. Chang-Hwan et al., 2017 [37], investigate a value-weighted classification approach
Sustainability 2023, 15, 15055 4 of 29

using an informational and theoretical filter approach. Cong Chen et al., 2016 [38], assert
that machine learning algorithms help us to analyze and avoid real-life damages. The
Naïve Bayes algorithm assesses the severity of driver injuries in rear-end vehicle crashes.
Ehab et al., 2018 [39], state that a smart grid involves the integration of various renewable
energy sources and has to be regulated effectively. The manuscript implements a modified
harmony search algorithm to show the efficiency of smart grids in terms of structure,
operation, and combined economic emission dispatch. Kadir Amasyali et al. [40] present a
thorough review of various energy consumption techniques and the results were discussed
to predict building energy consumption.
Demand response in buildings can be analyzed using reinforcement learning. Nima
Shiri et al., 2018 [41], utilize a Naïve Bayes-supervised machine learning algorithm in a
smart grid design that strategically controls and manages its resources based on condi-
tional probability. Abdorreza et al., 2018 [42], discuss the usage of modified imperialist
competitive algorithms for efficiently managing the energy resources in a smart grid. The
data mining methods are broadly divided into the following categories: database, sta-
tistical, machine learning methods, and neural network. Gengyuan Liu et al., 2018 [43],
highlights the developments in research in the fields of big data analytics and industrial
energy efficiency assessments, with a focus on the numerous energy efficiency techniques
which are based on process analysis of energy usage and big data mining. Muhammad
et al., 2018 [44], examine the usage of multi-agent systems in distributed smart grids to
strategically manage and control the energy resources which are involved. Panayiotis et al.,
2015 [45], present a novel internal energy balancing method using a decision tree machine
learning algorithm to address the increase and decrease in substantial load. Sook-Chin
et al., 2017 [46], assert that energy production in any sector must be generated at a good
level and maintained efficiently. Mark Landry et al., 2016 [47], utilize a Varrichio prob-
abilistic gradient boosting machine, asserting that it is an important asset in forecasting
wind energy production. Mehmet et al., 2017 [48], utilized real-time meteorological data
streaming to predict the power generated by wind using a KNN classifier algorithm. Raik
Becker et al., 2017 [49], utilized a hybrid approach, combining the k-nearest neighbors’
algorithm and the random forest algorithm, for forecasting the generation of wind power.
Simon Martinez et al., 2017 [50], propose that energy production can be carried out using
combined heat and power applications. This approach employs various machine learning
techniques to assess the energy output. Vehicle-to-grid transfer applications for electric
cars consists of distributed storage units, and their bidirectional features are surveyed
and analyzed. M. Pihlatie et al., 2014 [51], demonstrate a viable option using machine
learning techniques to study the practicality of running Nylund Fully Electric City Buses.
N. Shaukat et al., 2018 [52], propose a smart grid system that consists of the integration of
various renewable energy sources; they additionally provide a survey on electric vehicle
transportation using machine learning techniques. Jian Tang et al., 2023 [53], proposed
novel cooling techniques for a hybrid solar system, providing improved power control and
energy management techniques.
The performances of various machine learning algorithms for various energy grid
applications are recorded in the references mentioned above. There are different types
of algorithms that can be implemented in various energy grid applications [34]. A brief
explanation of algorithm types and the energy grid applications considered in this research
work are seen in Sections 2.2 and 2.3. The parameters considered and the implementations
of the machine learning algorithms differ in every paper; thus, an overall comparison of all
the available machine learning algorithms, using standard parameters, is essential. The
comparison study (Table 1) for all the machine learning algorithms accounted for here was
carried out considering standard parameters such as the energy grid application used, the
reasons for utilization, the pre-training requirements, and their objectives.
Sustainability 2023, 15, 15055 5 of 29

Table 1. Algorithm comparison.

Application Reason for Pre-

Ref. Year Learning Algorithm Objectives
Learning Usage Training
[1] 2014 Consumption ANN Forecasting Yes Energy storage
[2] 2004 Consumption General regression neural Cooling load Yes Energy cost
networks prediction
[3] 2008 Consumption Simulation tools Assessing Yes Peak demand
simulation tools
[4] 2008 Consumption ANN, energy plus Forecasting Yes Energy demand
[5] 2018 Consumption Multidimensional hybrid Forecasting Yes Comfort
integrated architecture
[6] 2008 Consumption ANN Forecasting Yes Energy needs
[7] 2016 Consumption SVR Forecasting Yes Occupancy
[8] 2013 Consumption ANN (upgraded) Forecasting Yes Time temperature Curve
[9] 2014 Consumption ANN Forecasting Yes Energy performance data
[10] 2012 Consumption ANN Analyze Yes Energy use
[11] 2013 Consumption SVR Forecasting Yes Energy conservation
[12] 2010 Consumption SVR Forecasting Yes Peak demand
[13] 2010 Consumption Fuzzy SVM and fuzzy C-mean Forecasting Yes Energy cost
clustering
[14] 2016 Consumption ANN, SVM, and ARIMA Forecasting Yes Energy consumption intensity
[15] 2018 Consumption Integrated BS and renewable Resource No Energy cost, GHGs reduction
energy sources management
[54] 2017 Consumption Classification, prediction, and Critical No Energy cost
reduction strategy assessment
[16] 2016 Consumption Echo state networks Forecasting, No Parameter sensitivity,
analysis energy cost
[17] 2011 Consumption New ANN Forecasting, Yes Customer flexibility,
validation energy cost
[18] 2011 Consumption Parallel support vector Forecasting, Yes Parallel computing,
machines analyzing energy data
[19] 2011 Consumption Hybrid genetic algorithm Forecasting Yes Parameter optimization
[20] 2013 Consumption Regression model Forecasting Yes Energy cost
[21] 2018 Consumption Pattern-recognition-based Forecasting No Improved prediction
algorithms performance
[22] 2005 Consumption Adaptive artificial neural Energy Yes Energy cost
networks prediction
[55] 2015 Consumption Multiple linear regression, Forecasting Yes Low computational cost
SVR, and ANN
[56] 2014 Consumption Neural network model Forecasting Yes Site load behaviors,
forecasting
[57] 2018 Consumption Gaussian process meta model Forecasting No Ship fuel consumption
[58] 2018 Consumption Data-driven techniques Forecasting Yes Identification of research gaps
[59] 2018 Consumption A hybrid artificial neural Forecasting Yes Convergence speed,
network optimization
[60] 2012 Consumption Indexed ARX model Forecasting Yes Energy cost, peak demand
[61] 2010 Consumption KPCA and SVM Forecasting No Energy cost, peak demand
Sustainability 2023, 15, 15055 6 of 29

Table 1. Cont.

Application Reason for Pre-

Ref. Year Learning Algorithm Objectives
Learning Usage Training
[62] 2009 Consumption Least square support vector Forecasting Yes Energy cost, peak demand
machine
[63] 2010 Consumption PCA and SVM Forecasting Yes Energy cost
[64] 2012 Consumption ANN Forecasting Yes Energy cost, peak demand
[65] 2014 Consumption Artificial neural network Forecasting Yes Energy cost
[66] 2018 Consumption SVM Forecasting Yes Energy cost
[67] 2018 Consumption Time series forecast techniques Forecasting Yes Quality analyzation
[68] 2009 Consumption Comparison SVM and ANN Forecasting Yes Energy cost, peak demand
[69] 2009 Consumption SVM Forecasting Yes Energy cost
[70] 2015 Consumption FFNN, RBFN, and ANFIS Forecasting No Energy cost, peak demand
[71] 2012 Consumption LS-SVM Forecasting Yes Energy cost, peak demand
[72] 2011 Consumption ANN-based MLP model Forecasting Yes Building occupancy and
demand
[73] 2014 Consumption SVR Forecasting Yes Energy cost, peak demand
[74] 2009 Consumption Neural network (modal Forecasting Yes Energy cost, peak demand
trimming method)
[75] 2015 Consumption Artificial neural network Forecasting Yes Energy demand
(ANN)
[76] 2010 Consumption ANN Analysis Yes Energy cost, peak demand
[77] 2016 Consumption Extreme learning machine Estimation Yes Energy demand
[78] 2017 Consumption Online Survey No Energy efficiency and
conservation
[79] 2014 Consumption Multi-model prediction and Forecasting Yes Energy demand
simulation
[80] 2018 Consumption Survey Estimation No Energy cost, peak demand
[81] 2017 Consumption ARIMA Forecasting Yes Energy demand
[82] 2011 Consumption Probabilistic entropy-based Forecasting Yes Building occupancy and
neural model demand
[83] 2015 Consumption Support vector machine Forecasting Yes Energy demand
prediction
[84] 2018 Consumption Research Benchmarking No Energy performance
[23] 2013 Consumption Multiple regression model Forecasting No Energy demand
[24] 2012 Consumption Using connectionist systems Forecasting No Energy demand, energy cost
[85] 2018 Consumption Survey Forecasting No Energy structure, energy usage
[25] 2018 Consumption Survey of data-driven Forecasting No Energy pattern profiling
techniques
[26] 2011 Consumption Short-term load forecasting Forecasting Yes Energy demand, energy cost
[86] 2011 Consumption Short-term load forecasting Forecasting Yes Demand flexibility, energy cost
[27] 2016 Consumption ANN Forecasting Yes Peak demand, energy cost
[28] 2014 Consumption Short-term forecasting Forecasting Yes Energy demand
[29] 2017 Consumption Survey Analysis No Energy conservation,
energy cost
Sustainability 2023, 15, 15055 7 of 29

Table 1. Cont.

Application Reason for Pre-

Ref. Year Learning Algorithm Objectives
Learning Usage Training
[30] 2016 Demand Reinforcement learning Forecasting Yes Cost reduction, smart charging
response algorithm
[87] 2015 Demand Reinforcement learning Evaluation No Numerical modeling,
response algorithm computational sustainability
[88] 2017 Demand Prediction-based multi-agent Prediction No Pattern change detection
response reinforcement learning
[89] 2016 Demand Reinforcement learning Evaluation No Sequential decision making,
response algorithm energy storage
[90] 2018 Demand Model-free control Evaluation Yes System identification
response scalability
[91] 2017 Demand Batch Reinforcement Learning Evaluation No Battery storage, optimization
response (Fitted-Q iteration algorithm)
[92] 2018 Demand Multiple agents and Evaluation No Convergence rate, reward
response reinforcement learning performance
[93] 2017 Demand Network-based Q-learning Analysis Yes Peak demand reduction,
response algorithm energy savings
[31] 2017 Demand Echo state network-based Analysis Yes Optimal battery control
response Q-learning method
[32] 2016 Demand Batch reinforcement learning Evaluation Yes Thermostatically controlled
response loads
[94] 2018 Demand Mobility-aware vehicle-to-grid Evaluation No Energy demand
response control algorithm (MACA)
[95] 2017 Demand Extended joint action learning Analysis No Demand flexibility
response
[33] 2018 Demand Deep reinforcement learning Analysis No Optimal control, energy
response efficiency
[96] 2018 Demand Reinforcement learning Evaluation No Thermal comfort
response
[34] 2017 Demand Reinforcement learning (used Evaluation Yes Thermal comfort, energy
response an ANN to map the storage
state–action)
[97] 2018 Demand Deep Q-learning-based Analysis Yes Energy storage, energy cost
response approach
[98] 2018 Demand Hierarchical reinforcement Prediction Yes Fuel cell performance,
response learning, upper confidence hydrogen consumption
tree search
[99] 2018 Demand Survey Analysis No Reliability; adaptability
response
[100] 2017 Demand Bayesian-regularized neural Evaluation Yes Optimization
response networks with genetic
algorithm,
reinforcement-learning-based
control logic using fitted
Q-iteration
[101] 2018 Demand Fuzzy Q-learning for Evaluation No Energy supply, energy cost,
response multi-agent decentralization reliability
[102] 2018 Demand Reinforcement learning-based Evaluation No Energy storage, power
response real-time power management management
Sustainability 2023, 15, 15055 8 of 29

Table 1. Cont.

Application Reason for Pre-

Ref. Year Learning Algorithm Objectives
Learning Usage Training
[103] 2006 Demand Reinforcement learning Analysis No Thermal storage, optimal
response control control
[104] 2018 Demand Online Markov chain-based Analysis No Hybrid tracked vehicle, fuel
response energy management economy
[35] 2017 Demand Deep transfer Q-learning with Analysis No Energy demand, energy cost
response virtual leader–follower
[36] 2018 Demand Real-time power management Evaluation No Energy storage, power
response strategy management
[105] 2015 Demand Centralized Lyapunov Analysis Yes Energy cost
response algorithm
[106] 2018 Demand Deep reinforcement learning Analysis Yes Fuel economy
response (DRL)
[107] 2018 Demand Reinforcement learning Analysis No Energy efficiency, thermal
response comfort
[108] 2015 Demand Device-based reinforcement Evaluation No Energy cost, energy demand
response learning
[37] 2018 Design Naïve Bayes Classification Yes Feature weighting, feature
selection
[38] 2016 Design Decision tree Naïve Bayes Prediction No Vehicle monitoring
(DTNB)
[39] 2018 Design Modified harmony search Analyzing No Economic emission dispatch
algorithm
[40] 2018 Design Survey Analyzing No Security, data quality
[109] 2018 Design Discrete wavelet transform Prediction No Wind speed intermittency
and Extreme learning machine
(DWT-ELM)
[41] 2018 Design Mixture of latent multinomial Prediction Yes Classification accuracy,
Naïve Bayes classifier conditional log-likelihood,
(MLMNB) under the ROC curve
[42] 2018 Energy Modified imperialist Evaluation No Operation cost, air pollution,
management competitive algorithm use of renewable energy
sources
[110] 2017 Energy Survey Evaluation No Energy constraint, use of
management harvested energy
[111] 2017 Energy Energy signature Evaluation No Energy use, peak demand
management
[43] 2018 Energy K-means clustering Evaluation No Energy efficiency, big data
management analysis
[112] 2018 Energy Non-linear multivariate Develop tool No Energy performance, energy
management regression model efficiency, energy cost
[113] 2016 Energy Survey Analysis No Power generation side
management management, smart grid and
renewable energy
management, asset
management and collaborative
operation, demand-side
management
[114] 2017 Energy Survey Analysis No Energy load, building
management operation, fraud detection
Sustainability 2023, 15, 15055 9 of 29

Table 1. Cont.

Application Reason for Pre-

Ref. Year Learning Algorithm Objectives
Learning Usage Training
[44] 2018 Energy Multi-agent systems Analysis No Optimal management
management
[45] 2016 Energy Decision tree Analysis No Storage planning, energy
management balancing
[46] 2017 Energy Linear Regression Evaluation No Detection of energy theft,
management defective smart meter
[115] 2018 Energy Equivalent consumption Evaluation No Minimization of fuel
management minimization strategy (ECMS), consumption, maintenance of
stochastic dynamic battery state of charge
programming (SDP)
[116] 2018 Production Survey Analysis No Energy demand, energy
recovery
[117] 2016 Production EnergyPlus simulation Evaluation No Incentive analysis
software
[118] 2018 Production K-means cluster Forecasting No Numerical weather prediction
[47] 2016 Production Probabilistic gradient boosting Forecasting No Wind track
machines for GEFCom2014
[48] 2017 Production kNN classifier Forecasting No Prediction accuracy
[49] 2017 Production Random forests Analysis No Selection of predictor variables
[119] 2018 Production Long short-term Forecasting No Peak load, frequency
memory-enhanced forget gate regulation
(LSTM-EFG)
[50] 2017 Production Survey Analysis No Energy demand, energy cost
[120] 2018 Production kNN method Evaluation No Fault Detection and Diagnosis
[121] 2018 Production Multiple imputation, Gaussian Forecasting Yes Dealing with missing data
Process Regression
[122] 2018 Production Survey Forecasting Yes Accurate forecasting,
optimization
[123] 2018 Storage Complex networks centrality Evaluation No Optimal positioning of storage
metrics systems
[124] 2018 Storage Survey Analysis No Flexibility
[125] 2016 V2g Survey Evaluation No Optimization
[126] 2018 V2g Survey Evaluation No Storage potential, connectivity
issues
[127] 2016 V2g Survey Evaluation Yes Remaining useful life
[51] 2014 V2g Survey Analysis No Lower total cost of ownership
[52] 2018 V2g Survey Analysis No Energy storage

The majority of currently used research methods are data-driven approaches; this
is considered a rapid advancement in information technology. Information assets are
considered as important to analyze, monitor, manage, improve, and predict energy in
industrial sectors. The energy management procedures must be improved because a
significant amount of revenue needs to be invested, and the investment that is carried
out must be profitable; all this means that it is important to choose a good and efficient
algorithm. The energy consumption patterns can be detected, operations can be optimized,
and maintenance of costs and energy demand prediction is observed in buildings [114].
Electricity is transmitted from generation end to consumers using an energy grid. In
Sustainability 2023, 15, 15055 10 of 29

industries energy grids are commonly used for various applications. Energy consumption
by buildings, production, vehicle-to-grid transfer, design, demand response, and energy
management are the energy grid applications considered in this research work. A thorough
analysis of the above-mentioned parameters was carried out using various algorithms.
Later, a comparative study is carried out and that gives us insights on the algorithms that
should be employed to meet our requirements. Ranking of the algorithms is carried out
to provide an overview of the parameters for which the machine learning algorithms are
highly efficient. Every algorithm has advantages of its own; the only concern is whether it
suits our chosen scenario, data type, and system requirements. Prediction performance is
an important indicator that is used to save energy and to minimize the revenue spent on
energy purchase and management. The wastage of energy can also be controlled through
this indicator.
The algorithms are ranked effectively which leads to various new queries, such as the
following: what data analysis tool should be used to evaluate the simulation results of the
top performing algorithms? There are numerous tools available for data analysis, but it
is important to understand and work on the tool that best suits a given scenario and its
system requirements. R-tool, Rapid Miner, and Weka are some of the leading data analysis
tools. A survey must be conducted to draw a conclusion on which data analysis tool must
be used for the performance evaluation. A scenario-oriented approach was carried out
using real-time data; the top-ranked algorithms were evaluated using the simulation results.
The obtained results are discussed here, and the ideal algorithm for energy management
application is determined [80].
In the industrial sector, energy-grid-based applications have remarkable potential is
not adequately valued. Recognizing novelty of prediction analysis frameworks is essential
in understanding energy consumption patterns, enabling one to control energy wastage and
energy costs efficiently. Furthermore, the implementation of the top-performing machine
learning algorithm may be the solution to managing and controlling energy flexibility,
demand, wastage, and cost in energy grid applications [34].
The proposed work in this article compares several machine learning algorithms
by taking essential parameters like consumption, production, vehicle-to-grid transfer,
storage, design, energy management, and demand response. The review is carried out
through a comparison among some key parameters, namely reasons for usage, objectives,
and pre-training. Following the comparison study, the various algorithms are further
graded with respect to the degree of efficiency of the parameters, namely large dataset,
speed, numeric prediction, dimension reduction, and the simplicity. Real-time data are
used to carry out a performance evaluation of the top-three-ranked algorithms for energy
management applications [100]. Results are concluded with the best-performing algorithm
being determined as the one with the highest accuracy and lowest error rates. This paper
highlights the comparative study of different machine learning algorithms for energy grid
parameters: consumption, production, energy management, demand response, design, and
vehicle-to-grid transfer. These analyses identify the gaps in smart grid implementation and
suggest future research guidelines for sustainability and energy optimization [34].
A wide range of machine learning algorithms for different energy grid applications
are available. The energy grid applications seen in our paper are consumption, production,
vehicle-to-grid transfer, storage, design, demand response, and energy management. All
possible algorithms for every application have different efficiencies for each parameter,
namely accuracy, linearity, self-training, and others. Here arises the need for a thorough
comparison analysis to rank the algorithms by considering deeper parameters and perfor-
mance evaluation in order to obtain the ideal algorithm amongst the various algorithms for
implementation in energy grid applications [34]. Numerous papers are taken as references
and insights are obtained; however, in the extensive comparison study, ranking and per-
formance evaluation of the machine learning algorithms were implemented in smart grid
systems; this fact is the driving factor of the work. Figure 1 provides an overview diagram
 each parameter, namely accuracy, linearity, self-training, and others. Here arises the need
for a thorough comparison analysis to rank the algorithms by considering deeper
parameters and performance evaluation in order to obtain the ideal algorithm amongst
the various algorithms for implementation in energy grid applications [34]. Numerous
papers are taken as references and insights are obtained; however, in the extensive
Sustainability 2023, 15, 15055 11 of 29
comparison study, ranking and performance evaluation of the machine learning
algorithms were implemented in smart grid systems; this fact is the driving factor of the
work. Figure 1 provides an overview diagram consisting of the relevant procedures,
consisting
which of the reviewing,
involves relevant procedures,
ranking, which involves reviewing,
and evaluating ranking,
the algorithms forand evaluating
energy grid
the algorithms for energy grid applications.
applications.

Figure 1. Overview
Figure 1. Overview diagram
diagramconsisting
consistingofof
thethe steps:
steps: reviewing,
reviewing, ranking,
ranking, and and evaluating
evaluating the
the algo-
algorithms for energy
rithms for energy gridgrid applications.
applications.

2. Algorithm
2. Algorithm Types
Types
Based on
Based onlabeled
labeledoror
unlabeled
unlabeleddatadata
types, the broad
types, classification
the broad categories
classification of machine
categories of
learning algorithms are supervised, unsupervised, and reinforcement learning.
machine learning algorithms are supervised, unsupervised, and reinforcement learning. Some
top-performing
Some and and
top-performing commonly
commonlyused machine
used machinelearning algorithms
learning algorithmsthatthatfall
fall under their
under their
respectivealgorithm
respective algorithmtypes
types are
are briefly
briefly explained
explained along
along with
with references.
references. Irrespective
Irrespective of theoftype
the
type
of of algorithm,
algorithm, theyto
they seem seem to beinused
be used in numerous
numerous energy
energy grid grid applications.
applications.

2.1. Supervised
2.1. Supervised Learning
Learning
The presence
The presence of
of aa supervisor
supervisor as
as aa guide
guideisisaabasic
basicexplanation
explanationofofsupervised
supervisedlearning.
learning.
Inferring an output from the trained labeled dataset that comprises a set of trained examples
Inferring an output from the trained labeled dataset that comprises a set of trained
is used in supervised learning. An input object in supervised learning is the learning
examples is used in supervised learning. An input object in supervised learning is the
dataset and the response obtained for the given raw data is the output value and is called a
learning dataset and the response obtained for the given raw data is the output value and
supervisory signal. The analysis of trained labeled dataset and production of an inferred
is called a supervisory signal. The analysis of trained labeled dataset and production of
function is observed using supervised learning algorithms and is used for mapping new
an inferred function is observed using supervised learning algorithms and is used for
examples with experiences obtained [118]. The class labels for the unseen instances are
mapping new examples with experiences obtained [118]. The class labels for the unseen
correctly determined for every algorithm in each optimal scenario. The learning algorithms
instances are correctly determined for every algorithm in each optimal scenario. The
identify a reasonable method to address unforeseen circumstances from the generalized
learning algorithms identify a reasonable method to address unforeseen circumstances
data. Concept learning also falls under supervised learning; it is a parallel task in animal
from the generalized data. Concept learning also falls under supervised learning; it is a
and human psychology. Linear regression is among the most commonly defined machine
parallel task in animal and human psychology. Linear regression is among the most
learning algorithms in supervised learning and can be used extensively in most hands-
commonly
on/real-time defined machine
applications; learning
this algorithms
is because in supervised
it is flexible learning
and can create linearand can beamong
relations used
unknown parameters to fit models which are non-linearly related to their parameters. The
statistical properties of subsequent estimators are used for identification of Naïve Bayes
classifiers in [46,67]. They were highly scalable and required a number of linear parameters;
this is because the number of variables in a learning algorithm and in the training proves is
determined by assessing the data to obtain its related output in a finite number of operations.
Assessing and training in closed-form expression takes time; rather, iterative approximation
as used for many other types of classifiers [37,38,41]. Support vector machine classifiers
perform data group classification and regression; these are mapped into that same space
and are predicted to belong to a given category based on the side of the gap which they
fall on [55,62,68]. The ANN techniques used in [22,59,62,79] are based on a collection of
connected units or nodes called artificial neurons, which loosely models the neurons in a
biological brain.
Sustainability 2023, 15, 15055 12 of 29

2.2. Unsupervised Learning

A machine is trained with data that are neither labeled nor classified, allowing the
machine learning algorithm to function without any instruction. Unlike the supervised
learning algorithm, no training is provided to the machine in unsupervised learning.
Grouping of unsorted data, recognizing its similarities and differences without any prior
instructions, is carried out by the machines. This learning is associated with learning with-
out a teacher; Hebbian learning is an example of unsupervised learning. The probability
densities of inputs are modeled, and self-organization is also seen in unsupervised learning.
There are various branches in machine learning; cluster analysis is one of the important
branches [55]. Grouping of data that do not have any categorization, classification, or
labeling is carried out in cluster analysis. Based on common properties, data were clustered
in [3,32,58] and the responses are based on the presence or absence of such commonalities.
Unsupervised learning encompasses sub-domains involving summarizing and explaining
data features, but the main central application is in the field of estimation of density. The
most common machine learning algorithms that are used in unsupervised learning are as
follows: K-means clustering is a method of vector quantization, and it is a commonly used
cluster analysis algorithm in data analysis. Numerous observations are partitioned into
some clusters, where each cluster split belongs to the nearest mean cluster [43,118]. Deep
belief network is a framework that can be implemented in machine learning algorithms
to process complex data inputs [8,11,19]. Principal-component-analysis-based variance
detection occurs when a vast majority of the data fall into a conventional distribution and
are sensitive to the relative scaling of the original variables [49,55,59].

2.3. Reinforcement Learning

The concept of reinforcement learning diverges from supervised learning; here, the
model is trained with the data that have the key answer within them. In reinforcement
learning, an agent takes the decision of the action that needs to be performed; when the
training set is found to be absent, the model learns by itself from previous experiences.
In [90,93], the utilized reinforcement learning approaches use software agents to take
necessary actions in a given environment to increase the notion of cumulative reward.
Statistics, information theory, simulation-based optimization, swarm intelligence, multi-
agent systems, control theory, and numerous problems from various fields are resolved due
to its generality. The utilization of dynamic programming techniques is seen in the Markov
decision process for the formulation of an environment [96,101]. In game theory and
economics, the equilibrium under bounded rationality is explained using reinforcement
learning. There are three machine learning paradigms mentioned in [104,106]: supervised,
unsupervised, and reinforcement learning. The focus on reinforcement learning is intended
to attain the performance which helps in finding the balance between current knowledge
(exploitation) and uncharted territory (exploration).

3. Energy Grid Applications

Vehicle-to-grid transfer, consumption, production, energy management, demand
response, and design are the energy grid applications considered in this study. A detailed
description of the usages, problems, and challenges faced during the processes involved in
energy grid applications are discussed below. A thorough comparative analysis of various
machine learning algorithms used in energy grid applications for different key parameters
shown in Figure 2 is carried out in Table 1.
 3. Energy Grid Applications
Vehicle-to-grid transfer, consumption, production, energy management, demand
response, and design are the energy grid applications considered in this study. A detailed
description of the usages, problems, and challenges faced during the processes involved
in energy grid applications are discussed below. A thorough comparative analysis of
Sustainability 2023, 15, 15055 13 of 29
various machine learning algorithms used in energy grid applications for different key
parameters shown in Figure 2 is carried out in Table 1.

Figure 2. Energy grid applications.

3.1. Demand Response

3.1. Demand Response
Demand
Demand response
response applications
applications are
are widely
widely used
used in
in energy
energy grids
grids and
and many
many studies
studies are
are
conducted and organized based on them. Demand response research and
conducted and organized based on them. Demand response research and study of generalstudy of general
information
information is is discussed
discussed and
and summarized
summarized in in this
this paper.
paper. InIn grid-connected
grid-connected systems,
systems, the
the
dissemination of power is a result of the due control on the demand side of
dissemination of power is a result of the due control on the demand side of electrical electrical con-
sumption.
consumption. Demand response
Demand research
response is reviewed
research [25] for[25]
is reviewed the study
for theof study
demand of response
demand
effects in residential houses; load shifting is determined, and its simulation results are then
response effects in residential houses; load shifting is determined, and its simulation
obtained. Therefore, load shifting is the ultimate use among demand response applica-
results are then obtained. Therefore, load shifting is the ultimate use among demand
tions [93,102]. Dynamic costs, incentives, and time-based demand responses comprise the
response applications [93,102]. Dynamic costs, incentives, and time-based demand
challenges faced in grid-connected system implementation scenarios. Simulation analyses
responses comprise the challenges faced in grid-connected system implementation
were carried out in [104,107] on the tariffs paid in controlling peak demands from electrical
scenarios. Simulation analyses were carried out in [104,107] on the tariffs paid in
appliances in houses; it was identified that the tariffs paid are high. Internet of Things
controlling peak demands from electrical appliances in houses; it was identified that the
applications and smart appliances present new views on energy consumption, production,
tariffs paid are high. Internet of Things applications and smart appliances present new
and management using demand response.
views on energy consumption, production, and management using demand response.
3.2. Energy Management
3.2. Energy Management
The behavior of energy storage and transportation of energy loads influences the
Theand
quality behavior
quantityof of
energy
energy storage
used in and transportation
buildings of energybasis;
on an everyday loadsthis
influences
can be seenthe
quality and quantity of energy used in buildings on an everyday basis;
in energy management systems. In the present context, energy usage in commercial and this can be seen in
energy management systems. In the present context, energy usage in commercial
residential buildings has significantly increased due to increasing populations [110,113]. and
residential
Energy buildings has
management significantly
systems increased
collect, store, due to increasing
and monitor the amount populations
of data that [110,113].
is avail-
Energy
able management
about energy use.systems
Analysiscollect, store, and monitor
and exploitation of data intheefficient
amountways of data that in
are seen is
available
this about energy
application. use. Analysis
Data analytics and exploitation
techniques of data
are presently in efficient
being ways areincrease
used to rapidly seen in
this application.
energy efficiency;Data
suchanalytics
researchtechniques
is receivingare presentlyinterest
significant being used to rapidly[114].
and attention increaseEn-
energy efficiency; such research is receiving significant interest and
ergy storage and controlling energy resources for energy balancing are problemsattention [114]. Energy
seen
storage
in energyand controllingapplications.
management energy resources for energy balancing
The communication networkare problems used
architectures seen in
energygrids
smart management applications.
are considered in [46]; The
here,communication
the intention was network architectures
to identify energy theftusedand in
smart grids
metering are considered
defects, aiming to in [46]; here,
decrease the the intention was
non-technical to that
losses identify
occurenergy theftgrids.
in smart and
A novel real-time energy management strategy is presented in [115]; this is proposed to
improve fuel consumption in hybrid vehicles through the utilization of different driving
strategies. Opportunities and challenges also arise in these techniques, leading to further
improvement requirements in computational technologies.

3.3. Energy Consumption

Practical data-driven models are commonly used in energy consumption applications,
especially for forecasting energy consumption [1,64,73]. In the past few years, with the
use of conventional sources and increase in demand requirements, energy consumption
and CO2 emissions have increased significantly. Energy is the most important part of
Sustainability 2023, 15, 15055 14 of 29

all our lives in the current context [17,23]. Data pre-processing is the most significant
method for energy prediction, and findings indicate that energy costs can be significantly
reduced. Dynamicity is the main problem seen in energy consumption and this can be
overcome using prediction analysis. The utilization of energy consumption prediction saves
energy cost and avoid wastage of excess energy [78]. Various machine learning algorithms
can be trained and tested to achieve the best results for energy consumption prediction.
Performance measures can be analyzed and evaluated using various data mining tools.
By developing and utilizing data-driven models, energy consumption prediction can be
improved in the near future [86]. Data-driven models can be used to remedy the existing
gaps in research fields and future for research.

3.4. Energy Production

Sources of primary energy are generated from electric power; this process is called
energy production. Delivering to the end users is the first stage followed by storage of
energy, recovery, transmission, and distribution. The significance of energy production is to
generate energy for various purposes, but it is commonly generated for industries. Electric
energy is not freely present in nature; therefore, it must be produced in remarkable amounts
via energy production [117]. Power plants and power stations generally carry out energy
production tasks. Electromechanical generators generate a huge amount of electricity using
power plants. Energy can be primarily produced through combustion or nuclear fission
methods and can also be produced through natural means; for example, kinetic energy can
be generated through freely available resources, such as wind or flowing water. Geothermal
and photovoltaic resources can also be used as energy sources. Various renewable and
non-renewable energy forms can be converted into useful electric energy [48,118]. Batteries
also provide a very small amount of utility in electric power. For utility-scale energy
generation, electric generators are rotated or photovoltaic systems are used [50,121]. The
main challenge seen in energy production applications is in smart grid scenarios, where
the production completely relies on solar panels; this approach is entirely dependent on
the weather and climate.

3.5. Design
Reduction in harmful gases and their emission must be controlled and, for this reason,
efficient models must be designed which address practical necessities, fault detections,
and feature weighting in designing any energy-related model. The main usage of design
modeling is to create ecofriendly and reliable models. There are various advancements
happening very rapidly in the energy field. Good designs are required which utilize
renewables-based distributed energy resources. Some examples for renewables-based
energy resources are wind and solar systems [38]. In smart grids, we see the concept of
active distribution level for the requirement of resilient power networks, and this can
be achieved using renewable-based distributed energy resources. Turbine technologies
have rapidly increased in the current context and concepts characterized by minimal land
requirements have also been formulated [38,39,109]. The main challenge involved in the
design of smart grids is the provision of suitable and safe protection approaches that involve
dynamic behavior with weather conditions. The other additional issues faced are mode
detection, varying fault scenarios, and section identification. Wind-turbine-based smart
grids are very commonly seen; these show the impacts in voltage–current characteristics
and consequently provide high wind speed profiles [41]. Pre-specified threshold settings
are not very sensitive for detecting the faults that could occur with varying wind speeds in
cases of conventional over-current relay scenarios.

3.6. Vehicle-to-Grid Transfer

Plug-in electric vehicles run on batteries, hydrogen fuel, or hybrid sources. These
electric vehicles communicate with the smart grid, relaying a supply–demand response
to the systems either by regulating their charge rate or by returning energy to the smart
Sustainability 2023, 15, 15055 15 of 29

grid. Distributed storage units are used in vehicle-to-grid technologies for electric cars. The
state of charge in batteries, technical data, and statistical data are seen in the power transfer
between vehicles and smart grids. Bidirectional power flow can be seen in vehicle-to-grid
transfers [51]. Power generation through wind and solar resources is commonly seen
in electric vehicles; in smart grids, under normal conditions, power is sent back to the
vehicle. The effect of intermittent energy supply is reduced using the distributed storage
units in electric vehicles. Efficient utilization of control schemes through optimal charging
and discharging is made as cost-effective as possible. The main usage of vehicle-to-grid
applications is to store and discharge energy. Intelligent scheduling for charging electric
vehicles is an emerging idea for obtaining maximum profits. Computer software is used to
analyze and find out the optimization in charging with and without vehicle-to-grid transfer.
Peak demand reduction is carried out, and the results show that better performance is
obtained through charging optimization with vehicle-to-grid than without vehicle-to-grid.
Vehicle-to-grid aggregators are introduced for providing additional frequency regulation
services due to rapid deployment of vehicle-to-grid technologies in electric vehicles. De-
mand from electric vehicle owners is fulfilled by using optimal dispatching strategies of
vehicle-to-grid aggregators [52]. The challenges faced in vehicle-to-grid applications are
battery degradation, investment costs, energy losses, and effects on distribution equipment.

4. Comparison Study
There are various algorithms used for each energy grid application and when it comes
to real-time implementations, there are numerous choices; further research is required in
this area. This research involves the comparison of the most frequently used algorithms
for the energy grid applications that are considered. The important parameters of ma-
chine learning algorithms—reasons for usage, pre-training involved, and objectives—are
compared for all the algorithms [58]. The reasons for usage of the machine learning al-
gorithms are mostly found to be forecasting, analysis, and evaluation, as seen in Table 1.
The pre-training parameter may or may not be mandatory for the machine learning algo-
rithms and that is also mentioned in the comparison table. Many papers and journals are
reviewed by taking into consideration the important parameters for every machine learn-
ing algorithm. The highlighted objectives for the energy grid applications are mentioned
clearly in the table. Energy costs, storage, and demand are some examples seen among the
objectives. This gives a clear idea to compare our requirements and analyze the existing
algorithms [34]. The energy grid applications are namely consumption, production, and
energy management, and vehicle-to-grid transfer, design, and demand response. This
chapter comprises various sections that involve reviewing, comparing, and ranking of
the various machine learning algorithms for each energy grid application. Comparison
tables considering the learning algorithm, reasons for usage, pre-training requirements, and
objectives for each paper reviewed is displayed in Table 1, and the key parameters—namely
accuracy, speed, linearity, training time, response time, self-learning, prediction numeric,
dimension reduction, simplicity and large datasets—are reviewed. Later, the compared
machine learning algorithms are ranked and the algorithms that excel are evaluated to
determine the ideal algorithm for energy management applications.

4.1. Comparison Table

A total of 71 algorithms are compared below. These algorithms were taken from
126 papers and reviewed carefully. All of these algorithms are used for energy grid applica-
tions, namely consumption, production, energy management, vehicle to grid applications,
storage, design, and demand response. Table 1 details the attributes considered here,
including application, learning algorithm, reasons for usage, pre-training, and objectives.
The comparisons require key parameters, as discussed from Section 4.1.1 to Section 4.1.10.
The weightage given to the parameters plays an important role in choosing the algorithm
that best suits each energy grid application. The top five algorithms for all the energy
grid applications—namely energy consumption, production, energy management, design,
Sustainability 2023, 15, 15055 16 of 29

demand response, and vehicle-to-grid transfer—are obtained with the help of comparison
carried out while considering deeper parameters. The top seven algorithms for energy grid
applications are detailed below.

4.1.1. Accuracy
The most accurate results are not always required in performance evaluation when
carried out using machine learning algorithms. Approximate answers are sufficient in
most evaluation cases, depending on system requirements. Processing time is drastically
reduced by using the most approximate method in machine learning algorithms [70,73].
The weightage given for the accuracy parameter must be high when compared to the
other parameters, since accuracy is the most important parameter considered. Acquiring
accuracy is the most challenging objective in the design of any algorithm [23,59]. Overfitting
is avoided naturally by most of the approximate methods. Accuracy is usually compared
and measured from many different sources among the collected data [26,60]. The algorithm
with the lowest accuracy is rejected and the one with highest accuracy will be chosen as the
optimal algorithm for any given energy grid application and scenario.

4.1.2. Speed
The time taken by the algorithm to run a complete analysis is defined as speed. It
is calculated in minutes, hours, seconds, or sometimes even in milliseconds. The faster
the algorithm runs, the better it is for our model [55]. We do not want to take a long time
for our process to run. Therefore, the speed parameter is considered to be very important.
The speed changes according to the density and size of a given dataset [60]. The machine
learning algorithms that are very slow are rejected and the ones that are performing with
high speed are used by the energy grid applications. The efficiency of any given machine
learning algorithm is determined by the speed it takes to perform the implementation [86].

4.1.3. Linearity
Linearity in the time variant system is maintained by different machine learning
algorithms. The classes are separated by straight lines in linear classification. Logistic
regression and support vector machines are example algorithms when linearity parameters
are considered. Regression algorithms presume that the datasets follow straight lines, and,
because of that, accuracy might be reduced. A high dimensional analog is comparable to
these desirable straight lines, and linearity is commonly represented in graphs [67,70].

4.1.4. Training Time

The time taken for training the data is considered to be very important; reduced
training time is desirable in this context. The training time taken for each machine learning
algorithm varies a great deal [60,65]. Accuracy is closely associated with the training time
and the sensitivity of some algorithms to the data points vary to the other data points.
Testing is carried out only after training the data. Most of the algorithms undergo a
lot of training, as it is beneficial for the algorithm in ensuring that it obtains improved
results [72,83]. This is also considered as a very important parameter when the algorithms
are compared.

4.1.5. Time Response

The time response is the time taken by any model to respond to a given situation or
any circumstance. The response time must be low so that a lot of time can be saved [60]. The
faster the response time, the more likely the algorithm is to be used. Successful computing
of any machine learning algorithm becomes critical if it has low response time [82]. The
elapsed time between the query and response in any given system is called the response
time [86]. System performance is measured using the total responses it gives based on
service requests.
Sustainability 2023, 15, 15055 17 of 29

4.1.6. Self-Learning
The self-learning parameter is the most important feature in the current context
because it allows the algorithm to perform tasks by itself from previous observations.
Artificial intelligence decides on its own to perform a task, without an algorithm prompt. In
order to decide and act by itself, the machine learning algorithm goes through an elaborate
training process [118]. The training process involves significant human input along with
observed values [8]. Specific situations are provided in advance and a lot of training is
carried out [19]. The problem is defined accurately, and the correct and incorrect possible
answers are also uploaded. The training data are then labeled, and the correct and incorrect
answers are evaluated [22]. Now, the algorithm knows how to react to each of the situations.
This is also known as a self-adaptive feature.

4.1.7. Predicting Numbers

The predicting numeric process usually happens in a stream and the next value must
be predicted [17]. There are various examples seen for this parameter, namely artificial
neural networks. This is also one of the important parameters that needs to be considered
when comparing algorithms [25]. Numerical predictions are the foundation for designing
any analytical approach to predict values. A predictive numeric is performed after refining
and evaluating the models that were trained [102]. Numerical prediction is evaluated and
assessed in all the machine learning algorithms.

4.1.8. Dimensions Reduction

The dimension reduction parameter is mainly seen in the statistics domain. Obtaining
some principle variables through reducing the number of random variables is the process
of dimension reduction [60]. The dataset taken can consist of many numbers of columns
with a three-dimensional space [90]. The process of dimension reduction involves bringing
down multiple columns to a very low number of columns within a two-dimensional space.
This can be divided into feature extraction and selection [46].

4.1.9. Simplicity
The machine learning algorithm is always fed with dataset for analysis and evaluation.
Some datasets will have numerous rows and columns while the other datasets will have
fewer rows and columns. Numerous steps were involved in every machine learning algo-
rithm to perform calculation and analysis; some being simple while the other algorithms are
complex [38,46]. Simple algorithms are considered easy for use while complex algorithms
seem to yield more accurate results [45].

4.1.10. Dataset
When the dataset is fed to every algorithm, it consists of various rows and columns,
in which 80% of the dataset is utilized for training and 20% of the dataset is used for
testing. Depending on the size of the dataset, it can be named large or small [60]. When
a larger dataset is applied for training, more accurate results will be obtained [62,75]. A
few thousand lines of rows are considered to be a large dataset, and the machine learning
algorithm takes more time to produce the results, leading to increase in runtime [118].
Large datasets can be used for scenarios and system requirements where the runtime is less
of a priority compared to the accuracy.

4.2. Energy Grid Applications and Machine Learning Algorithms

The reason for usage of every energy grid application is unique and serves different
purposes. In Table 1, the reasons for the usage of energy grid applications are mentioned,
and the majority are forecasting, analyzing, and evaluating. Most of the dataset needs
pre-training, as mentioned in Table 1. The comparison table shows that pre-training
helps us to attain better results. The default split between training and testing is seen
to be 80:20. First, the dataset is trained for the algorithm with 80% of the dataset, and
Sustainability 2023, 15, 15055 18 of 29

then the algorithm is tested with the remaining 20% of the dataset. The objective of
every scenario is unique; a reduction in energy costs is the main goal of this research
work [2,15–17,54,65,66]. Peak demands can be forecasted in advance to counterbalance the
smart grid scenario [12,62,64,70–73]. Energy wastage can be highly controlled; revenue can
be saved in a large margin through prediction [30,108,112,116]. These are the two main
objectives considered here.
The algorithm we selected to use for the scenario plays a pivotal role in decision
making. Survey, comparison, and performance evaluation of the algorithms will help us in
determining which algorithm would suit the present smart grid scenario and the chosen
requirements better. The ANN algorithm was used in most of the scenarios where the
system’s learning and prediction was a priority [1,6,14,27,64,65,75,76]. SVM analyzes the
data for classification and regression analysis [14,61,66,69]. The linear regression algorithm
shows that the level of predictability seems to have a lower margin of error [25,29,46,99].
An interactive Q-learning algorithm learns the strategy and instructs the representative
to take necessary action under specific conditions [30,31,89,93,97,100,101]. The decision
tree algorithm is used for predicting multiple variables [46,58,85,99]. The essences of the
machine learning algorithms presented in the comparison table are utilized for ranking
them with respect to the energy grid application; the top-performing algorithms are noted
for further performance evaluation. Ranking of the machine learning algorithms according
to certain parameters—large dataset, speed, dimension reduction, predicting numeric, and
simplicity—possess different efficiency levels, and this gives insights into which algorithm
can be deployed for a given dataset, scenario, and requirement.

5. Results and Discussions

In this section, we discuss the seven top-performing algorithms based on ranking
the machine learning algorithms designed for energy grid applications. The parameters
used for ranking the algorithms are large dataset, speed, predicting numeric, dimension
reduction, and simplicity.
Sustainability 2023, 15, x FOR PEER REVIEW 20 of 31
The comparative study of various algorithms in the energy grid system on demand
response, consumption, production, energy management, design, and vehicle-to-grid
transfer provides the details of greater insight on top seven performing algorithms as seen
in Figure 3. Figure
applications. 4 shows
The overall the
top overall
seven top seven algorithms
best-performing for the
algorithms are energy grid applications.
determined, irrespective
The overall
of the top seven
energy best-performing
grid applications. algorithms
Deeper are determined,
parameters consideredirrespective of the energy
for the ranking of the
grid applications. Deeper parameters considered for the ranking of the algorithms
algorithms are size of dataset, speed, predicting numeric values, and dimension simplicity are size
of dataset,
[87]. speed, predicting numeric values, and dimension simplicity [87].

Figure 3. Workflow
Workflow of
of comparison
comparison study.
study.
Sustainability 2023, 15, 15055 19 of 29
Figure 3. Workflow of comparison study.

Figure 4. Overall top seven algorithms for the energy grid applications.

A
A clear
clear understanding
understanding of of the
the efficiency
efficiency levels
levels of the seven
of the seven top-performing
top-performing algorithms
algorithms
are seen (see Figure 4, overall top seven algorithms for the energy grid applications).
are seen (see Figure 4, overall top seven algorithms for the energy grid applications). Every
Every algorithm
algorithm has different
has different efficiency
efficiency levelseach
levels under under each parameter
parameter and the variances
and the variances are seen
are seen
in the in the
graph thatgraph
assiststhat assists
us in choosingus in
thechoosing
preferablethealgorithm
preferableforalgorithm
the given for the given
requirements
requirements
and scenario. and scenario.
As the result As
of the
the result
ranking of the
hasranking has been
been carried out,carried
we seeout, wethe
that seealgo-
that
the algorithms—namely Naïve Bayes, artificial neural networks, linear regression,
rithms—namely Naïve Bayes, artificial neural networks, linear regression, support vector support
vector
machine,machine, Q-learning,
Q-learning, GaussianGaussian
mixture mixture
model,model, and principle
and principle component
component analysis—
analysis—were
were compared [58]. A brief explanation of the distinct features of these
compared [58]. A brief explanation of the distinct features of these machine learning machine learning
algo-
algorithms can be seen below. This explanation further guides us in understanding
rithms can be seen below. This explanation further guides us in understanding the algo- the
algorithm that would best suit our dataset and requirements.
rithm that would best suit our dataset and requirements.
5.1. Comparison Ranking of Top Five Algorithms for Each Energy Grid Applications
5.1. Comparison Ranking of Top Five Algorithms for Each Energy Grid Applications
Accuracy, training time, reaction time, linearity, and self-learning are the primary com-
Accuracy,
parative metricstraining time, reaction
used to choose time,
the top five linearity, for
algorithms and self-learning
each energy gridare the primary
application [6].
comparative metrics used to choose the top five algorithms for each energy
Now, a deeper comparison of the top five algorithms for each energy grid application grid applica-
tion
is [6]. Now,
carried a deeper
out with morecomparison of the top they
intense parameters; five algorithms for each
are then further energy
ranked to grid appli-
obtain the
cation is carried out with more intense parameters; they are then further ranked
best-performing algorithm for each energy grid application: demand response, energy to obtain
the best-performing
management, algorithm
consumption, for each energy
production, grid
design, andapplication: demand response, energy
vehicle-to-grid.
management, consumption, production, design, and vehicle-to-grid.
5.1.1. Production
5.1.1.Delivering
Productionto the end users is the first stage followed by storage of energy, recovery,
Delivering
transmission, to the end etc.
distribution, users is the energy
Electric first stage followed
is not by storage
freely available in of energy,
nature, andrecovery,
it has to
transmission,
be generated indistribution,
remarkable etc. Electric
amounts energyenergy
through is not production.
freely available in nature,
Power andpower
plants and it has
to be generated
stations carry outinenergy
remarkable amounts
production through
tasks. energy production.
Electromechanical Power
generators plantshuge
generate and
amounts of electricity through the use of power plants [48,118]. Energy can be primarily
produced using combustion or nuclear fission methods, and can also be produced by
other means; for example, energy can be produced through kinetic energy, which is freely
available in natural resources such as the wind or flowing water. Geothermal and photo-
voltaic resources can also be used as energy sources. Various renewable and non-renewable
energy forms can be converted into useful electric energy. Figure 5 shows a comparison of
the algorithms for production applications; the random decision forest was found to be
correct—decision trees have a habit of overfitting to their training datasets [120,121]. The
figure was obtained through a ranking of the efficiency levels of the parameters of the top
five algorithms for the production applications.
 freely available in natural resources such as the wind or flowing water. Geothermal and
photovoltaic resources can also be used as energy sources. Various renewable and non-
renewable energy forms can be converted into useful electric energy. Figure 5 shows a
comparison of the algorithms for production applications; the random decision forest was
Sustainability 2023, 15, 15055
found to be correct—decision trees have a habit of overfitting to their training datasets
20 of 29
[120,121]. The figure was obtained through a ranking of the efficiency levels of the param-
eters of the top five algorithms for the production applications.

Figure5.5.Comparison
Figure Comparisonof
ofalgorithms
algorithmsfor
forproduction
productionapplications.
applications.

5.1.2.
5.1.2. Consumption
Consumption
Figure
Figure 66 shows
shows aa comparison
comparison of of various
various algorithms
algorithms in in order
order to to predict
predict the
the energy
energy
utilization
utilizationand andconsumption
consumption application.
application. TheThe graph
graph shown
shown in in Figure
Figure 66shows
showsthe thesubset
subset
clustering
clustering as a function of consumption and is carried out based on the estimationindex
as a function of consumption and is carried out based on the estimation index
of
of aa user’s energy
energy consumption;
consumption;the therecorded
recordedindexindexofof various
various users’
users’ power
power consump-
consumption
tion patterns
patterns werewere obtained.
obtained. Practical
Practical data-driven
data-driven modelsmodels are commonly
are commonly seenseen in energy
in energy con-
consumption applications,
sumption applications, especially
especially forfor energy
energy consumption
consumption prediction
prediction [1,64,73].
[1,64,73]. Due Due to
to in-
increases
creases ininpopulations,
populations,there
thereisisan
anincreasing
increasingdemand
demandfor for energy,
energy, and
and usage
usage ofof conven-
conven-
tional
tional sources
sources increases
increases CO
CO22 emissions.
emissions. Energy
Energy is is the
the most
most essential
essential resource
resource in in all
all our
our
lives in the current context. Data pre-processing is an important method
lives in the current context. Data pre-processing is an important method used in energy used in energy
prediction;
prediction;according
according totothe
theprediction
prediction results,
results, energy
energy costs
costs can
can bebe reduced.
reduced. Various
Various ma- ma-
chine learning algorithms can be trained and tested to ensure the best energy
chine learning algorithms can be trained and tested to ensure the best energy consumption consumption
prediction
predictionresults.
results.The
Theperformance
performancemeasures
measures cancanbebe
analyzed
analyzed andand
evaluated
evaluated using various
using vari-
Sustainability 2023, 15, x FOR PEER REVIEW
data mining tools [10,13,14]. The figure was obtained by ranking the efficiency levels 22
of of
the31
ous data mining tools [10,13,14]. The figure was obtained by ranking the efficiency levels
parameters for the for
of the parameters topthe
fivetop
algorithms for consumption
five algorithms applications.
for consumption applications.

Figure6.6.Comparison
Figure Comparisonofofalgorithms
algorithmsfor
forconsumption
consumptionapplications.
applications.

5.1.3.
5.1.3.Vehicle-to-Grid
Vehicle-to-GridTransfer
Transfer
Figure
Figure77shows
showsthe thegraph
graphofofalgorithm
algorithmcomparison
comparisonfor foraavehicle-to-grid
vehicle-to-gridapplication
application
that
thatshows
showsgreater
greaterinsights
insightsinto
intogenetic
geneticalgorithms
algorithmsthat thatare
arecommonly
commonlyusedusedtotogenerate
generate
superior solutions for system optimization and other search problems. Bidirectional
superior solutions for system optimization and other search problems. Bidirectional power
flows can be seen in vehicle-to-grid transfers [51]. Efficient utilization of control schemes
power flows can be seen in vehicle-to-grid transfers [51]. Efficient utilization of control in
optimal charging and discharging becomes as cost-effective as possible with advancements
schemes in optimal charging and discharging becomes as cost-effective as possible with
made in vehicle-to-grid
advancements made intechnology. Intelligent
vehicle-to-grid scheduling
technology. for charging
Intelligent electric
scheduling for vehicles
charging
iselectric
an emerging
vehiclesapproach to ensure
is an emerging maximum
approach profits
to ensure are obtained.
maximum profitsComputer software
are obtained. Com-
puter software is used for analyses to determine the optimization in charging with and
without vehicle-to-grid transfer. Peak demand reduction is carried out, and the results
show that better performance is obtained when charging optimization with vehicle-to-
grid transfer than without vehicle-to-grid transfer [52]. The figure was obtained by rank-
 Figure 7 shows the graph of algorithm comparison for a vehicle-to-grid application
that shows greater insights into genetic algorithms that are commonly used to generate
superior solutions for system optimization and other search problems. Bidirectional
power flows can be seen in vehicle-to-grid transfers [51]. Efficient utilization of control
schemes in optimal charging and discharging becomes as cost-effective as possible with
Sustainability 2023, 15, 15055 21 of 29
advancements made in vehicle-to-grid technology. Intelligent scheduling for charging
electric vehicles is an emerging approach to ensure maximum profits are obtained. Com-
puter software is used for analyses to determine the optimization in charging with and
iswithout
used forvehicle-to-grid
analyses to determine
transfer. the optimization
Peak in charging
demand reduction with and
is carried out,without
and thevehicle-
results
to-grid
show that better performance is obtained when charging optimization with that
transfer. Peak demand reduction is carried out, and the results show better
vehicle-to-
performance is obtained when charging optimization with vehicle-to-grid transfer
grid transfer than without vehicle-to-grid transfer [52]. The figure was obtained by rank- than
without vehicle-to-grid transfer [52]. The figure was obtained by ranking the efficiency
ing the efficiency levels of the parameters of the top five algorithms for vehicle-to-grid
levels of the parameters of the top five algorithms for vehicle-to-grid applications.
applications.

Figure7.7.Comparison
Figure Comparisonof
ofalgorithms
algorithmsfor
forvehicle-to-grid
vehicle-to-gridapplications.
applications.

5.1.4.
5.1.4.Design
Design
The
Thedesign
designmodel
modelusesusesBayesian
Bayesianmethods
methodsas asititprevents
preventsoverfitting;
overfitting;ititisisecofriendly
ecofriendly
and
andreliable.
reliable.There
Therearearevarious
variousadvancements
advancementshappening
happeningvery veryrapidly
rapidly in
in the
the energy
energyfield;
field;
their
theireffective
effectivedesigns
designsareareurgently
urgentlyrequired
required for for
useuse
in renewable-energy-based
in renewable-energy-based distributed
distrib-
systems. This can
uted systems. becan
This achieved using renewable-energy-based
be achieved using renewable-energy-based distributed energy energy
distributed resources.
re-
Use
sources. Use of turbine technology has rapidly increased in the current context andwith
of turbine technology has rapidly increased in the current context and concepts con-
Sustainability 2023, 15, x FOR PEER REVIEW
minimal land requirements have also been 23 of 31
cepts with minimal land requirements haveproposed
also been[38,39].
proposedThe[38,39].
main challenge
The main involved
challenge
in smart grid
involved design
in smart gridis design
providing suitable suitable
is providing and safeand protection that involves
safe protection dynamic
that involves dy-
behaviors in response to weather conditions. Figure 8 shows a comparison
namic behaviors in response to weather conditions. Figure 8 shows a comparison of the of the algorithms
under design
algorithms applications;
under the figure was
design applications; obtained
the figure wasby ranking
obtained bythe efficiency
ranking levels of lev-
the efficiency the
parameters of the top of
els of the parameters five
thealgorithms for designfor
top five algorithms applications.
design applications.

Figure8.8. Comparison
Figure Comparisonof
ofalgorithms
algorithmsfor
fordesign
designapplications.
applications.

5.1.5.
5.1.5. Energy
Energy Management
Management
The
The decisionforests
decision forestsalgorithm
algorithmandandthe
thelinear regression
linear algorithm
regression algorithmareare
seen to excel
seen in
to excel
the parameters comparison when compared with the other top algorithms
in the parameters comparison when compared with the other top algorithms under en- under energy
management
ergy managementapplications. The efficient
applications. analysis
The efficient and exploitation
analysis of data
and exploitation are seen
of data in this
are seen in
application [99,100].
this application In thisInpaper,
[99,100]. a smart
this paper, grid scenario
a smart is takenisinto
grid scenario consideration
taken for the
into consideration
energy
for the management application
energy management to analyze
application usage patterns
to analyze among among
usage patterns users and employees
users and em-
ployees in the building sector. Opportunities and challenges also arise in the techniques
used, which lead to further improvement requirements in computational technologies.
Figure 9 presents a comparison of algorithms under the energy management application;
this figure was obtained by ranking the efficiency levels of the parameters for the top five
 Figure 8. Comparison of algorithms for design applications.

5.1.5. Energy Management

The decision forests algorithm and the linear regression algorithm are seen to excel
in the parameters comparison when compared with the other top algorithms under en-
Sustainability 2023, 15, 15055 22 of 29
ergy management applications. The efficient analysis and exploitation of data are seen in
this application [99,100]. In this paper, a smart grid scenario is taken into consideration
for the energy management application to analyze usage patterns among users and em-
in the building
ployees sector. Opportunities
in the building and challenges
sector. Opportunities also arise
and challenges alsoinarise
the techniques used,
in the techniques
which lead to further improvement requirements in computational technologies.
used, which lead to further improvement requirements in computational technologies. Figure 9
presents a comparison of algorithms under the energy management application;
Figure 9 presents a comparison of algorithms under the energy management application; this figure
was
this obtained
figure wasbyobtained
ranking bytheranking
efficiency
thelevels of thelevels
efficiency parameters for the top five
of the parameters algorithms
for the top five
under energy management applications.
algorithms under energy management applications.

Figure9.
Figure 9. Comparison
Comparison of
of algorithms
algorithms for
forenergy
energymanagement
managementapplications.
applications.

5.1.6.
5.1.6. Demand
Demand Response
Response
Demand
Demandresponse
responseresearch
researchandandstudy
studyofof general
generalinformation
information is observed
is observed andand
summa-
sum-
rized
marizedin this paper.
in this Under
paper. the smart
Under grid concept,
the smart the dissemination
grid concept, of demand
the dissemination response
of demand re-
applications was influenced
sponse applications by the idea
was influenced by theofidea
controlling the demand
of controlling side of
the demand electrical
side con-
of electrical
sumption.
consumption. DemandDemandresponse research
response is reviewed
research in [25].
is reviewed LoadLoad
in [25]. shifting is used
shifting in residen-
is used in res-
tial houses for the study of demand response effects; then, simulation results
idential houses for the study of demand response effects; then, simulation results are ob- are obtained.
Sustainability 2023, 15, x FOR PEER REVIEW
Controlling the peakthe
tained. Controlling demand of the electrical
peak demand appliances
of the electrical in houses
appliances where where
in houses 24 ofare
high tariffs
high 31
tar-
paid must be evaluated and understood using simulation studies. Figure
iffs are paid must be evaluated and understood using simulation studies. Figure 10 presents 10 presents a
comparison
a comparison ofofalgorithms
algorithmsunder
underdemand
demandresponse
responseapplications
applications[31–33].
[31–33]. TheThe figure
figure was
was
obtained by ranking the efficiency levels of the parameters of the top five
obtained by ranking the efficiency levels of the parameters of the top five algorithms for algorithms for
demand response applications.
demand response applications.

Figure 10. Comparison of algorithms for demand response applications.

Figure 10. Comparison of algorithms for demand response applications.

5.2. Comparison Ranking of Top Three Algorithms for Each Energy

Energy Management
Management Applications
Applications
The top
topthree
threebest-performing
best-performing algorithms—linear
algorithms—linear regression,
regression, artificial
artificial neural
neural net-
networks,
works, and decision
and decision tree—have
tree—have been determined
been determined undermanagement
under energy energy management applications
applications (Figure 9
(Figure 9 presents a comparison of the algorithms for energy management applications).
These are evaluated and the results are discussed to find the ideal algorithm for the given
smart grid scenario. The machine learning algorithms chosen for use in a grid-connected
system, along with their analyses using data analysis tools (Weka, R-Tool, and Rapid
Sustainability 2023, 15, 15055 23 of 29

presents a comparison of the algorithms for energy management applications). These are
evaluated and the results are discussed to find the ideal algorithm for the given smart
grid scenario. The machine learning algorithms chosen for use in a grid-connected system,
along with their analyses using data analysis tools (Weka, R-Tool, and Rapid Miner), are
shown in Table 2.

Table 2. Survey of data mining tools.

Data Analysis Tool Weka Rapid Miner R-Tool

Language Java Language-independent C, Fortran, R
Image identification at the grid point,
Statistical analysis utilized to
Flexible to use and can be statistical analysis and attribute
Advantages take major decision in
extended in Rapid Miner selection, and detection of parameter
grid-connected system
for system optimization
Poor documentation, weak
Less specialized for data
classical statistics, poor Requires detailed knowledge of
Limitations mining, requires prominent
parameter optimization, weak database handling
knowledge of array language
csv reader
Statistical analysis, data mining,
Type Machine learning Statistical computing
predictive analysis
It is best suited for mining Specialized for business solutions that It has a large number of users
Specialization association rules and data include predictive analysis and in the fields of bioinformatics
mining techniques statistical computing and social science

The three top-ranked algorithms for the energy management application need to be
further evaluated to assess the simulation results and to find the ideal algorithm that would
best suit the smart grid system. Various data mining tools are currently available and the
top data mining tools and their features are compared. Each data mining tool has its own
pros and cons.
The data mining tool that best suits the smart grid scenario and needs is chosen. The
leading data mining tools—Weka, Rapid Miner, and R tool—are compared in detail [5].
Key features, like languages, are used to build the tool. The advantages of the specific data
mining tools, the various limitations of each tool, the type of data mining tool that each one
is, and the specialization of each data mining tool are discussed. The main features required
for the scenario is predictive analysis; Rapid Miner is optimal for this in comparison with
the other tools. R tool is the topmost data mining tool, but this type is mainly involved with
statistical computing. The Rapid Miner tool is the best-performing data analysis tool for
the given smart grid scenario and system requirements. Performance evaluation is carried
out for the top three algorithms—linear regression, artificial neural networks, and decision
tree—for the energy management application and the results are compared to find the ideal
algorithm which has a high accuracy rate and low error rates.

5.3. Discussion on Performance Evaluation

The three different algorithms—decision tree, linear regression, and neural networks—do
not have significant differences, and they all perform well. Python is used in the Rapid
Miner tool to obtain the true computation of the performances of the three top-ranked
algorithms for energy management applications. Linear regression works very well with
consistent data. Decision tree and neural networks also seem to have a low margin of error,
but we see that linear regression has the lowest occurrence of errors. Neural networks
perform well for scenarios where they must learn and predict well; decision tree is good for
predicting multiple variables, which is not required in this smart grid scenario. The analysis
of the performance evaluation of the data shows linear correlation between every single
event and is systematically aligned to a consumption range at any given point. Hence,
Sustainability 2023, 15, 15055 24 of 29

linear regression works best in our given smart grid scenario with real-time data for energy
management applications.

6. Conclusions
The goal of this study was to find out which algorithm would help us predict better
outputs for consumption. A deep and complete comparative analysis of the possible
algorithms used for different energy grid applications were reviewed. Comparison of the
various machine learning algorithms that can be used to yield better results was conducted
to narrow the pool, and further comparisons with deeper parameters were carried out. In
this research work, a review of recent research developments regarding the use of machine
learning algorithms for applications is carried out. Analyses of 126 research papers for
energy-related applications, namely storage, vehicle-to-grid transfer, energy management,
design, consumption, production, and demand response, were carried out. Overall, the
top seven algorithms of all the applications were compared. A total of 71 algorithms were
compared, taking a few parameters into consideration, namely training time, accuracy,
response time, linearity, and self-learning.
Energy management systems play major roles in tariff reduction and in maintaining
sustainability in grid-connected systems. This can be attained with the use of demand-
in-response algorithms by integrating concepts of cyberattacks and smart meters. The
main challenges in grid-connected systems are their vulnerability to energy theft and
faults. Advanced smart control strategies can be incorporated to ensure better communi-
cation strategies and reliable integration. Among various machine learning techniques,
the seven algorithms (Naïve Bayes, artificial neural networks, linear regression, support
vector machine, Q-learning, Gaussian mixture model, and principle component analy-
sis) were selected, as they provide various advantages for the development of intelligent
power technologies. Forecasting energy utilization tends to increase gradually and the
accumulated datasets from various buildings are large. The evaluation indexes for energy
consumption characteristics are divided into groups and are assessed using mutual infor-
mation matrices; total energy consumption is predicted based on multiple linear regression
algorithms. Real-time energy theft is an issue both at the supply section and the consumer
section; providing system sustainability can be attained using hybrid techniques using
SVM and deep learning. The uses of this hybrid technique are as follows: (1) analyze and
pre-process historical data on energy usage; (2) train and test datasets to prepare for energy
theft; (3) extract valuable data from smart meters and classify the energy used with the
application of Naïve Bayes; (4) validate the proposed model. The Gaussian mixture model
was one of the seven algorithms chosen because it is flexible and can accomplish hard
clustering for complex data. Based on the outcome of the review, a selection of the best
algorithms that suit each application, in terms of their different characteristics, was made.
This selection is presented in the Results and Discussion Sections. The research concludes
with the results of the algorithms evaluated; our results determined that linear regression
is the optimal choice for our scenario.

Author Contributions: Methodology, R.K., S.N. and Y.T.; Software, R.K., S.N. and Y.T.; Formal
analysis, R.K., S.N. and Y.T.; Investigation, R.K., S.N. and Y.T.; Resources, R.K., S.N. and Y.T.;
Writing—original draft, R.K., S.N. and Y.T.; Writing—review and editing, R.K., S.N. and Y.T.; Supervi-
sion, R.K., S.N. and Y.T.; Project administration, R.K.; Funding acquisition, S.N. and Y.T. All authors
have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data presented in this study are available in the main text of
the article.
Sustainability 2023, 15, 15055 25 of 29

Acknowledgments: The team of authors acknowledges anonymous reviewers for their feedback,
which certainly improved the clarity and quality of this paper.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Bagnasco, A.; Fresi, F.; Saviozzi, M.; Silvestro, F.; Vinci, A. Electrical consumption forecasting in hospital facilities: An application
case. Energy Build. 2015, 103, 261–270. [CrossRef]
2. Ben-Nakhi, A.E.; Mahmoud, M.A. Cooling load prediction for buildings using general regression neural networks. Energy
Convers. Manag. 2004, 45, 2127–2141. [CrossRef]
3. Yezioro, A.; Dong, B.; Leite, F. An applied artificial intelligence approach towards assessing building performance simulation
tools. Energy Build. 2008, 40, 612–620. [CrossRef]
4. Neto, A.H.; Fiorelli, A.S. Comparison between detailed model simulation and artificial neural network for forecasting building
energy consumption. Energy Build. 2008, 40, 2169–2176. [CrossRef]
5. Matéa, A.; Peral, J.; Ferrández, A.; Trujillo, D.G.J. A hybrid integrated architecture for energy consumption prediction. Future
Gener. Comput. Syst. 2016, 63, 131–147. [CrossRef]
6. Ekici, B.B.; Aksoy, U.T. Prediction of building energy consumption by using artificial neural networks. Adv. Eng. Softw. 2009, 40,
356–362. [CrossRef]
7. Massana, J.; Pous, C.; Burgas, L.; Melendez, J.; Colomer, J. Short-term load forecasting for non- residential buildings contrasting
artificial occupancy attributes. Energy Build. 2016, 130, 519–531. [CrossRef]
8. Roldán-Blay, C.; Escrivá-Escrivá, G.; Álvarez-Bel, C.; Roldán-Porta, C.; Rodríguez-García, J. Upgrade of an artificial neural
network prediction method for electrical consumption forecasting using an hourly temperature curve model. Energy Build. 2013,
60, 38–46. [CrossRef]
9. Scarlat, N.; Dallemand, J.; Turhana, C.; Kazanasmazb, T.; Uygunb, I.E.; Ekmena, K.E.; Akkurta, G.G. Comparative study of a
building energy performance software(KEP-IYTE-ESS) and ANN-based building heat load estimation. Energy Build. 2014, 85,
115–125. [CrossRef]
10. Hawkins, D.; Hong, S.M.; Raslan, R.; Mumovic, D.; Hanna, S. Determinants of energy use in United Kingdom higher education
buildings using statistical and artificial neural network methods. Int. J. Sustain. Built Environ. 2012, 1, 50–63. [CrossRef]
11. Liu, D.; Chen, Q. Prediction of Building Lighting Energy Consumption Based on Support Vector Regression algorithm. In Pro-
ceedings of the 2013 9th Asian Control Conference (ASCC), Istanbul, Turkey, 23–26 June 2013. [CrossRef]
12. Solomon, D.; Wintery, R.; Boulangerz, A.; Andersonz, R.; Wu, L. Forecasting Energy Demand in Large Commercial Buildings Using
Support Vector Machine Regression; Department of Computer Science, Columbia University: New York, NY, USA, 2012. [CrossRef]
13. Li, X.; Deng, Y.; Ding, L.; Zhong, J.L. Building Cooling Load Forecasting Using Fuzzy Support Vector Machine and Fuzzy C-Mean
Clustering. In Proceedings of the 2010 International Conference on Computer and Communication Technologies in Agriculture
Engineering, Chengdu, China, 12–13 June 2010. [CrossRef]
14. Zhao, D.; Zhong, M.; Zhang, X.; Su, X. Energy consumption predicting model of VRV (Variable refrigerant volume) system in
office buildings based on data mining. Energy 2016, 102, 660–668. [CrossRef]
15. Ahmed, F.; Naeem, M.; Ejaz, W.; Iqbal, M.; Anpalagan, M.A. Resource management in cellular base stations powered by renewable
energy sources. J. Netw. Comput. Appl. 2018, 112, 1–17. [CrossRef]
16. Shi, G.; Liu, D.; Wei, Q. Energy consumption prediction of office buildings based on echo state networks. Neuro Comput. 2016, 216,
478–488. [CrossRef]
17. Escrivá, G.E.; Álvarez-Bel, C.; Blay, C.R.; Ortega, M.A. New artificial neural network prediction method for electrical consumption
forecasting based on building end-uses. Energy Build. 2011, 43, 3112–3119. [CrossRef]
18. Zhao, H.X.; Magoules, F. Parallel Support Vector Machines Applied to the Prediction of Multiple Buildings Energy Consumption.
J. Algorithms Comput. Technol. 2010, 4, 231–249. [CrossRef]
19. Li, K.; Sua, H.; Chua, J. Forecasting building energy consumption using neural networks and hybrid neuro-fuzzy system: A
comparative study. Energy Build. 2011, 43, 2893–2899. [CrossRef]
20. Korolija, I.; Zhang, Y.; Halburd, L.M.; Hanby, V.I. Regression models for predicting United Kingdom office building energy
consumption from heating and cooling demands. Energy Build. 2013, 59, 214–227. [CrossRef]
21. Wang, J.; Li, G.; Chen, H.; Liu, J.; Guo, Y.; Sun, S.; Hu, Y. Energy consumption prediction for water-source heat pump system
using pattern recognition-based algorithms. Appl. Therm. Eng. 2018, 136, 755–766. [CrossRef]
22. Yang, J.; Rivard, H.; Zmeureanu, R. On-line building energy prediction using adaptive artificial neural networks. Energy Build.
2005, 37, 1250–1259. [CrossRef]
23. Catalina, T.; Iordache, V.; Caracaleanu, B. Multiple regression model for fast prediction of the heating energy demand. Energy
Build. 2013, 57, 302–312. [CrossRef]
24. Kamaev, V.A.; Shcherbakov, M.V.; Panchenko, D.P.; Shcherbakova, N.L.; Brebels, A. Using Connectionist Systems for Electric
Energy Consumption Forecasting in Shopping Centers. Upr. Bol’shimi Sist. 2010, 31, 92–109. [CrossRef]
25. Wei, Y.; Zhang, X.X.; Shi, Y.; Xia, L.; Pan, S.; Wu, J.; Han, M.; Zhao, X. A review of data-driven approaches for prediction and
classification of building energy consumption. Renew. Sustain. Energy Rev. 2018, 82, 1027–1047. [CrossRef]
Sustainability 2023, 15, 15055 26 of 29

26. Penya, Y.K.; Borges, C.E.; Agote, D.; Fernández, I. Short-term load forecasting in air-conditioned non- residential Buildings. In
Proceedings of the 2011 IEEE International Symposium on Industrial Electronics, Gdansk, Poland, 27–30 June 2011. [CrossRef]
27. Chae, Y.T.; Horesh, R.; Hwang, Y.D.; Lee, Y.M. Artificial neural network model for forecasting sub- hourly electricity usage in
commercial buildings. Energy Build. 2016, 111, 184–194. [CrossRef]
28. Iwafune, Y.; Ikegami, T. Short-term Forecasting of Residential Building Load for Distributed Energy Management. In Proceedings
of the ENERGYCON 2014, Cavtat, Croatia, 13–16 May 2014. [CrossRef]
29. Gao, Z.; Li, Y.; Ning, Y. The survey and analysis on the energy consumption of hospital buildings in Shandong province. Procedia
Eng. 2017, 205, 2485–2492. [CrossRef]
30. Chis, A.; Lund, J.; Koivunen, V. Reinforcement Learning-Based Plug-in Electric Vehicle Charging with Forecasted Price. IEEE
Trans. Veh. Technol. 2017, 66, 3674–3684. [CrossRef]
31. Shi, G.; Liu, D.; Wei, Q. Echo state network-based Q-learning method for optimal battery control of offices combined with
renewable energy. IET Control Theory Appl. 2017, 11, 915–922. [CrossRef]
32. Ruelens, F.; Claessens, B.J.; Vandael, S.; De Schutter, B.; Babuška, R.; Belmans, R. Residential Demand Response of Thermostatically
Controlled Loads Using Batch Reinforcement Learning. IEEE Trans. Smart Grid 2017, 8, 2149–2159. [CrossRef]
33. Kazmi, H.; Mehmood, F.; Lodeweyckx, S.; Driesen, J. Giga watt hour scale savings on a budget of zero: Deep reinforcement
learning based optimal control of hot water systems. Energy 2018, 144, 159–168. [CrossRef]
34. Vázquez-Canteli, J.; Kämpf, J.; Nagy, Z. Balancing comfort and energy consumption of a heat pump using Batch Reinforcement
learning with fitted Q-iteration. Energy Procedia 2017, 122, 415–420. [CrossRef]
35. Zhang, X.; Bao, T.; Yu, T.; Yang, B.; Han, C. Deep transfer Q-learning with virtual leader-follower for supply demand Stackelberg
game of smart grid. Energy 2017, 133, 348–365. [CrossRef]
36. Xiong, R.; Duan, Y.; Cao, J.; Yu, Q. Battery and ultra capacitor in-the-loop approach to validate a real-time power management
method for an all-climate electric vehicle. Appl. Energy 2018, 217, 153–165. [CrossRef]
37. Lee, C.-H. An information-theoretic filter approach for value weighted classification learning in naive Bayes. Data Knowl. Eng.
2018, 113, 116–128. [CrossRef]
38. Chen, C.; Zhang, G.; Yang, J.; Milton, J.C.; Alcántara, A. An explanatory analysis of driver injury severity in rear-end crashes
using a decision table/Naïve Bayes (DTNB) hybrid classifier. Accid. Anal. Prev. 2016, 90, 95–107. [CrossRef] [PubMed]
39. Elattar, E.E. Modified harmony search algorithm for combined economic emission dispatch of smart grid incorporating renewable
sources. Energy 2018, 159, 496–507. [CrossRef]
40. Ghofrani, M.; Steeble, A.; Barrett, C.; Daneshnia, I. Survey of Big Data Role in Smart Grids: Definitions, Applications, Challenges,
and Solutions. Open Electr. Electron. Eng. J. 2018, 12, 86–97. [CrossRef]
41. Harzevili, N.S.; Alizadeh, S.H. Mixture of latent multinomial naive Bayes classifier. Appl. Soft Comput. 2018, 69, 516–527.
[CrossRef]
42. Rabiee, A.; Sadeghi, M.; Aghaei, J. Modified imperialist competitive algorithm for environmental constrained energy management
of smart grids. J. Clean. Prod. 2018, 202, 273–292. [CrossRef]
43. Liu, G.; Yang, J.; Hao, Y.; Zhang, Y. Big data-informed energy efficiency assessment of China industry sectors based on K-means
clustering. J. Clean. Prod. 2018, 183, 304–314. [CrossRef]
44. Khan, M.W.; Wang, J.; Ma, M.; Xiong, L.; Li, P.; Wu, F. Optimal energy management and control aspects of distributed smart grid
using multi-agent systems. Sustain. Cities Soc. 2019, 44, 855–870. [CrossRef]
45. Moutis, P.; Kazakos, S.S.; Brucoli, M. Decision tree aided planning and energy balancing of planned community smart grids. Appl.
Energy 2016, 161, 197–205. [CrossRef]
46. Yip, S.C.; Wong, K.S.; Hewa, W.P.; Ga, M.-T.; Phan, R.C.-W.; Tan, S.-W. Detection of energy theft and defective smart meters in
smart grids using linear regression. Electr. Power Energy Syst. 2017, 91, 230–240. [CrossRef]
47. Landry, M.; Erlinger, T.P. Probabilistic gradient boosting machines for GEFCom2014 wind forecasting. Int. J. Forecast. 2016, 32,
1061–1066. [CrossRef]
48. Yesilbudak, M.; Sagiroglu, S.; Colak, I. A novel implementation of kNN classifier based on multi tupled meteorological input data
for wind power prediction. Energy Convers. Manag. 2017, 135, 434. [CrossRef]
49. Becker, R.; Thrän, D. Completion of wind turbine data sets for wind integration studies applying random forests and k-nearest
neighbors. Appl. Energy 2017, 208, 252–262. [CrossRef]
50. Martinez, S.; Michaux, G.; Patrick, S.; Bouvier, J.-L. Smart-combined heat and power systems (smart-CHP) based on renewable
energy sources. Energy Convers. Manag. 2017, 154, 262–285. [CrossRef]
51. Pihlatie, M.; Kukkonen, S.; Halmeaho, T.; Karvonen, V.; Nylund, N.-O. Fully Electric City Buses The Viable Option. In Proceedings
of the 2014 IEEE International Electric Vehicle Conference (IEVC), Florence, Italy, 17–19 December 2014. [CrossRef]
52. Shaukat, N.; Khan, B.; Ali, S.M.; Mehmood, C.A.; Khan, J.; Farid, U.; Majid, M.; Anwar, S.M.; Jawad, M.; Ullah, Z. A survey on
electric vehicle transportation within smart grid system. Renew. Sustain. Energy Rev. 2018, 81, 1329–1349. [CrossRef]
53. Tang, J.; Ni, H.; Peng, R.-L.; Wang, N.; Zuo, L. A review on energy conversion using hybrid photovoltaic and thermoelectric
systems. J. Power Sources 2023, 562, 232785. [CrossRef]
54. Zhao, G.Y.; Liu, Z.Y.; He, Y.; Cao, H.J.; Guo, Y.B. Energy consumption in machining: Classification, prediction, and reduction
strategy. Energy 2017, 133, 142–157. [CrossRef]
Sustainability 2023, 15, 15055 27 of 29

55. Massana, J.; Pous, C.; Burgas, L.; Melendez, J.; Colomer, J. Short-term load forecasting in a non- residential building contrasting
models and attributes. Energy Build. 2015, 92, 322–330. [CrossRef]
56. Jetcheva, J.G.; Majidpour, M.; Chen, W.-P. Neural network model ensembles for building-level electricity load forecasts. Energy
Build. 2014, 84, 214–223. [CrossRef]
57. Yuan, J.; Nian, V. Ship Energy Consumption Prediction with Gaussian Process Meta model. Energy Proceedia 2018, 152, 655–660.
[CrossRef]
58. Amasyali, K.; El-Gohary, N.M. A review of data-driven building energy consumption prediction studies. Renew. Sustain. Energy
Rev. 2018, 81, 1192–1205. [CrossRef]
59. Li, K.; Xie, X.; Xue, W.; Dai, X.; Chen, X.; Yang, X. A hybrid teaching-learning artificial neural network for building electrical
energy consumption prediction. Energy Build. 2018, 174, 323–334. [CrossRef]
60. Yun, K.; Luck, R.; Mago, P.J.; Cho, H. Building hourly thermal load prediction using an indexed ARX model. Energy Build. 2012,
54, 225–233. [CrossRef]
61. Li, X.; Ding, L.; Lǔ, J.; Xu, G.; Li, J. A Novel Hybrid Approach of KPCA and SVM for Building Cooling Load Prediction.
In Proceedings of the 2010 Third International Conference on Knowledge Discovery and Data Mining Symposium, Washington,
DC, USA, 25–28 July 2010. [CrossRef]
62. Li, X.; Lu, J.; Ding, L.; Xu, G.; Li, J. Building Cooling Load Forecasting Model Based on LS-SVM. In Proceedings of the 2009
Asia-Pacific Conference on Information Processing, Shenzhen, China, 18–19 July 2009. [CrossRef]
63. Lv, J.; Li, X.; Ding, L.; Zhong, J.L. Applying Principal Component Analysis and Weighted Support Vector Machine in Building
Cooling Load Forecasting. In Proceedings of the 2010 International Conference on Computer and Communication Technologies
in Agriculture Engineering, Chengdu, China, 12–13 June 2010. [CrossRef]
64. Leung, M.C.; Tse, N.C.F.; Lai, L.L.; Chow, T.T. The use of occupancy space electrical power demand in building cooling load
prediction. Energy Build. 2012, 55, 151–163. [CrossRef]
65. Paudel, S.; Elmtiri, M.; Kling, W.L.; Le Corre, O.; Lacarrière, B. Pseudo dynamic transitional modeling of building heating energy
demand using artificial neural network. Dep. Energy Build. 2014, 70, 81–93. [CrossRef]
66. Amasyali, K.; El-Gohary, N. Building lighting energy consumption prediction for supporting energy data analytics. Procedia Eng.
2016, 145, 511–517. [CrossRef]
67. Wang, Q.; Li, S.; Li, R. Forecasting energy demand in China and India: Using single-linear, hybrid-linear, and non-linear time
series forecast techniques. Energy 2018, 161, 821–831. [CrossRef]
68. Li, Q.; Meng, Q.; Cai, J.; Yoshino, H.; Mochida, A. Predicting hourly cooling load in the building: A comparison of support vector
machine and different artificial neural networks. Energy Convers. Manag. 2009, 50, 90–96. [CrossRef]
69. Li, Q.; Meng, Q.; Cai, J.; Yoshino, H.; Mochida, A. Applying support vector machine to predict hourly cooling load in the building.
Appl. Energy 2009, 86, 2249–2256. [CrossRef]
70. Jovanović, R.Ž.; Sretenović, A.A.; Živković, B.D. Ensemble of various neural networks for prediction of heating energy consump-
tion. Energy Build. 2015, 94, 189–199. [CrossRef]
71. Edwards, R.E.; New, J.; Parker, L.E. Predicting future hourly residential electrical consumption: A machine learning case study.
Energy Build. 2012, 49, 591–603. [CrossRef]
72. Kwok, S.S.; Yuen, R.K.; Lee, E.W. An intelligent approach to assessing the effect of building occupancy on building cooling load
prediction. Build. Environ. 2011, 46, 1681–1690. [CrossRef]
73. Jain, R.K.; Smith, K.M.; Culligan, P.J.; Taylor, J.E. Forecasting energy consumption of multi-family residential buildings using
support vector regression: Investigating the impact of temporal and spatial monitoring granularity on performance accuracy.
Appl. Energy 2014, 123, 168–178. [CrossRef]
74. Yokoyama, R.; Wakui, T.; Satake, R. Prediction of energy demands using neural network with model identification by global
optimization. Energy Convers. Manag. 2009, 50, 319–327. [CrossRef]
75. Ferlito, S.; Atrigna, M.; Graditi, G.; De Vito, S.; Salvato, M.; Buonanno, A.; Di Francia, G. Predictive models for building’s energy
consumption: An Artificial Neural Network (ANN) approach. In Proceedings of the 2015 XVIII AISEM Annual Conference,
Trento, Italy, 3–5 February 2015. [CrossRef]
76. Wong, S.L.; Wan, K.K.W.; Lam, T.N.T. Artificial neural networks for energy analysis of office buildings with day lighting. Appl.
Energy 2010, 87, 551–557. [CrossRef]
77. Naji, S.; Keivani, A.; Shamshirband, S.; Alengaram, U.J.; Jumaat, M.Z.; Mansor, Z.; Lee, M. Estimating building energy
consumption using extreme learning machine method. Energy Syst. 2016, 97, 506–516. [CrossRef]
78. Hu, S.; Yan, D.; Guo, S.; Cui, Y.; Dong, B. A survey on energy consumption and energy usage behavior of households and
residential building in urban China. Energy Build. 2017, 148, 366–378. [CrossRef]
79. Farzana, S.; Liu, M.; Baldwin, A.; Hossain, M.U. Multi-model prediction and simulation of residential building energy in urban
areas of Chongqing, South West China. Energy Build. Syst. 2014, 81, 161–169. [CrossRef]
80. Al Qadi, S.; Sodagar, B.; Elnokaly, A. Estimating the heating energy consumption of the residential buildings in Hebron, Palestine.
J. Clean. Prod. 2018, 196, 1292–1305. [CrossRef]
81. Li, S.; Li, R. Comparison of Forecasting Energy Consumption in Shandong, China Using the ARIMA Model, GM Model, and
ARIMA-GM Model. Sustainability 2017, 9, 1181. [CrossRef]
Sustainability 2023, 15, 15055 28 of 29

82. Kwok, S.S.K.; Lee, E.W.M. A study of the importance of occupancy to building cooling load in prediction by intelligent approach.
Energy Convers. Manag. 2011, 52, 2555–2564. [CrossRef]
83. Paudel, S.; Nguyen, P.H.; Kling, W.L.; Elmitri, M.; Lacarrière, B.; Le Corre, O. Support Vector Machine in Prediction of Building
Energy Demand Using Pseudo Dynamic Approach Proceedings of ECOS 2015. In Proceedings of the 28th International Conference
on Efficiency, Cost, Optimization, Simulation and Environmental Impact of Energy Systems (ECOS 2015)—Palais Beaumont, Pau,
France, 30 June–3 July 2015.
84. Sun, D. Research and Application of Energy Consumption Benchmarking Method for Public Buildings Based on Actual Energy
Consumption. Energy Procedia 2018, 152, 475–483. [CrossRef]
85. Xu, J.; Gao, W.; Huo, X. Analysis on energy consumption of rural building based on survey in northern China. Energy Sustain.
Dev. 2018, 47, 34–38. [CrossRef]
86. Penya, Y.K.; Borges, C.E.; Fernández, I. Short term load forecasting in non-residential Buildings. In Proceedings of the IEEE
Africon 2011 Symposium on Industrial Electronics, Victoria Falls, Zambia, 13–15 September 2011. [CrossRef]
87. de Gracia, A.; Fernández, C.; Castell, A.; Mateu, C.; Cabeza, L.F. Control of a PCM ventilated facade using reinforcement learning
techniques. Energy Build. 2015, 106, 234–242. [CrossRef]
88. Marinescu, A.; Dusparic, I.; Clarke, N. Prediction-Based Multi-Agent Reinforcement Learning in Inherently Non-Stationary
Environments. ACM Trans. Autom. Adapt. Syst. 2017, 12, 23. [CrossRef]
89. Ruelens, F.; Claessens, B.J.; Quaiyum, S.; De Schutter, B.; Babu, R.; Belmans, R. Reinforcement Learning Applied to an Electric
Water Heater: From Theory to Practice. IEEE Trans. Smart Grid 2018, 9, 3792–3800. [CrossRef]
90. Claessens, B.J.; Vanhoudt, D.; Desmedt, J.; Ruelens, F. Model-free control of thermostatically controlled loads connected to a
district heating network. Energy Build. 2018, 159, 1–10. [CrossRef]
91. Mbuwir, B.V.; Ruelens, F.; Spiessens, F.; Deconinck, G. Battery Energy Management in a Smart grid Using Batch Reinforcement
Learning Brida V. Mbuwir, Frederik Ruelens, Fred Spiessens, Geert Deconinck. Energies 2017, 10, 1846. [CrossRef]
92. Jiang, C.X.; Jing, Z.X.; Cui, X.R.; Ji, T.Y.; Wu, Q.H. Multiple agents and reinforcement learning for modeling charging loads of
electric taxis. Appl. Energy 2018, 222, 158–168. [CrossRef]
93. Mahapatra, C.; Moharana, A.K.; Leung, V.C.M. Energy Management in Smart Cities Based on Internet of Things: Peak Demand
Reduction and Energy Savings. Sensors 2017, 17, 2812. [CrossRef] [PubMed]
94. Ko, H.; Pack, S. Mobility-Aware Vehicle-to-Grid Control Algorithm in Smart grids. IEEE Trans. Intell. Transp. Syst. 2018, 19,
2165–2174. [CrossRef]
95. Hurtado, L.A.; Mocanu, E.; Nguyen, P.H.; Gibescu, M.; Kamphuis, I.G. Enabling Cooperative Behavior for Building Demand
Response Based on Extended Joint Action Learning. IEEE Trans. Ind. Inform. 2018, 14, 127–136. [CrossRef]
96. Brusey, J.; Hintea, D.; Gaura, E.; Beloe, N. Reinforcement learning-based thermal comfort control for vehicle cabins. Mechatronics
2018, 50, 413–421. [CrossRef]
97. Wu, J.; He, H.; Peng, J.; Li, Y.; Li, Z. Continuous reinforcement learning of energy management with deep Q network for a power
split hybrid electric bus. Appl. Energy 2018, 222, 799–811. [CrossRef]
98. Yuan, J.; Yang, L.; Chen, Q. Intelligent energy management strategy based on hierarchical approximate global optimization for
plug-in fuel cell hybrid electric vehicles. Int. J. Hydrogen Energy 2018, 43, 8063–8078. [CrossRef]
99. Vázquez-Canteli, J.R.; Nagy, Z. Reinforcement learning for demand response: A review of algorithms and modeling techniques.
Appl. Energy 2019, 235, 1072–1089. [CrossRef]
100. Schmidt, M.; Moreno, M.V.; Schülke, A.; Macek, K.; Pastor, A.G. Optimizing legacy building operation: The evolution into
data-driven predictive cyber-physical systems. Energy Build. 2017, 148, 257–279. [CrossRef]
101. Kofinas, P.; Dounis, A.I.; Vouros, G.A. Fuzzy Q-Learning for multi-agent decentralized energy management in smart grids. Appl.
Energy 2018, 219, 53–67. [CrossRef]
102. Xiong, R.; Cao, J.; Yu, Q. Reinforcement learning-based real-time power management for hybrid energy storage system in the
plug-in hybrid electric vehicle. Appl. Energy 2018, 211, 538–548. [CrossRef]
103. Liu, S.; Henze, G.P. Experimental analysis of simulated reinforcement learning control for active and passive building thermal
storage inventory. Energy Build. 2006, 38, 142–147. [CrossRef]
104. Liu, T.; Wang, B.; Yang, C.L. Online Markov Chain-based energy management for a hybrid tracked vehicle with speedy Q-learning.
Energy 2018, 160, 544–555. [CrossRef]
105. Liu, Y.; Yuen, C.; Hassan, N.U.; Huang, S.; Yu, R.; Xie, S. Electricity Cost Minimization for a Smart grid with Distributed Energy
Resource under Different Information Availability. IEEE Trans. Ind. Electron. 2015, 62, 2571–2583. [CrossRef]
106. Hu, Y.; Li, W.; Xu, K.; Zahid, T.; Qin, F.; Li, C. Energy Management Strategy for a Hybrid Electric Vehicle Based on Deep
Reinforcement Learning. Appl. Sci. 2018, 8, 187. [CrossRef]
107. Chen, Y.; Norford, L.K.; Samuelson, H.W.; Malkawi, A. Optimal control of HVAC and window systems for natural ventilation
through reinforcement learning. Energy Build. 2018, 169, 195–205. [CrossRef]
108. Wen, Z.; O’Neill, D.; Maei, H. Optimal Demand Response Using Device-Based Reinforcement Learning. IEEE Trans. Smart Grid
2015, 6, 2312–2324. [CrossRef]
109. Manohar, M.; Koley, E.; Ghosh, S. Smart grid protection under wind speed intermittency using extreme learning machine. Comput.
Electr. Eng. 2018, 72, 369–382. [CrossRef]
Sustainability 2023, 15, 15055 29 of 29

110. Babayo, A.A.; Anisi, M.H.; Ali, I. A Review on energy management schemes in energy harvesting wireless sensor networks.
Renew. Sustain. Energy Rev. 2017, 76, 1176–1184. [CrossRef]
111. Arregi, B.; Garay, R. Regression Analysis of the energy consumption of tertiary buildings. Energy Procedia 2017, 122, 9–14.
[CrossRef]
112. Bilous, I.; Deshko, V.; Sukhodub, I. Parametric analysis of external and internal factors influence on building energy performance
using non-linear multivariate regression models. J. Build. Eng. 2018, 20, 327–336. [CrossRef]
113. Zhou, K.; Fu, C.; Yang, S. Big data driven smart energy management: From big data To big insights. Renew. Sustain. Energy Rev.
2016, 56, 215–225. [CrossRef]
114. Molina-Solana, M.; Rosa, M.; Ruiz, M.D.; Gómez-Romero, J.; Martin-Bautista, M.J. Data science for building energy management:
A review. Renew. Sustain. Energy Rev. 2017, 70, 598–609. [CrossRef]
115. Jiao, X.H.; Li, Y.; Xu, F.; Jing, Y. Real-time energy management based on ECMS with stochastic optimized adaptive sytem
equivalence factor for HEVs. Cogent Eng. 2018, 5, 1540027. [CrossRef]
116. Di Maria, F.; Sisani, F.; Contini, S. Are EU waste-to-energy technologies effective for exploiting the energy in bio-waste? Appl.
Energy 2018, 230, 1557–1572. [CrossRef]
117. Zhang, J.; Cho, H.; Knizley, A. Evaluation of financial incentives for combined heat and power (CHP) systems in U.S. regions.
Renew. Sustain. Energy Rev. 2016, 59, 738–762. [CrossRef]
118. Wang, K.; Qi, X.; Liu, H.; Song, J. Deep belief network based k-means cluster approach for short-term wind power forecasting.
Energy 2018, 165, 840–852. [CrossRef]
119. Yu, R.; Gao, J.; Yu, M.; Lu, W.; Xu, T.; Zhao, M.; Zhang, J.; Zhang, R.; Zhang, Z. LSTM-EFG for wind power forecasting based on
sequential correlation features. Future Gener. Comput. Syst. 2019, 93, 33–42. [CrossRef]
120. Madeti, S.R.; Singh, S.N. Modeling of PV system based on experimental data for fault detection using kNN method. Sol. Energy
2018, 173, 139–151. [CrossRef]
121. Liu, T.; Wei, H.; Zhang, K. Wind power prediction with missing data using Gaussian process regression and multiple imputation.
Appl. Soft Comput. 2018, 71, 905–916. [CrossRef]
122. Das, U.K.; Tey, K.S.; Seyedmahmoudian, M.; Mekhilef, S.; Idris, M.Y.I.; Van Deventer, W.; Horan, B.; Stojcevski, A. Forecasting of
photovoltaic power generation and model optimization: A review. Renew. Sustain. Energy Rev. 2018, 81, 912–928. [CrossRef]
123. Korjani, S.; Facchini, A.; Mureddu, M.; Caldarelli, G.; Damiano, A. Optimal positioning of storage systems in smart grids based
on complex networks centrality measures. Sci. Rep. 2018, 8, 16658. [CrossRef] [PubMed]
124. Paiho, S.; Saastamoinen, H.; Hakkarainen, E.; Similä, L.; Pasonen, R.; Ikäheimo, J.; Rämä, M.; Tuovinen, M.; Horsmanheimo, S.
Increasing flexibility of Finnish energy systems—A review of potential technologies and means. Sustain. Cities Soc. 2018, 43,
509–523. [CrossRef]
125. Tan, K.M.; Ramachandaramurthy, V.K.; Yong, J.Y. Integration of electric vehicles in smart grid: A review on vehicle to grid
technologies and optimization techniques. Renew. Sustain. Energy Rev. 2016, 53, 720–732. [CrossRef]
126. Mahmud, K.; Town, G.E.; Morsalin, S.; Hossain, M.J. Integration of electric vehicles and management in the internet of energy.
Renew. Sustain. Energy Rev. 2018, 82, 4179–4203. [CrossRef]
127. Wu, L.; Fu, X.; Guan, Y. Review of the Remaining Useful Life Prognostics of Vehicle Lithium-Ion Batteries Using Data-Driven
Methodologies. Appl. Sci. 2016, 6, 166. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.