Energy Reports 12 (2024) 3654–3670

Contents lists available at ScienceDirect

Energy Reports
journal homepage: www.elsevier.com/locate/egyr

Review article

Review on smart grid load forecasting for smart energy management using
machine learning and deep learning techniques
Biswajit Biswal a , Subhasish Deb a,* , Subir Datta a, Taha Selim Ustun b, Umit Cali c,d,**
a
Department of Electrical Engineering, Mizoram University, Aizawl, Mizoram 796004, India
b
Fukushima Renewable Energy Institute, AIST (FREA), Koriyama 963-0298, Japan
c
Norwegian University of Science and Technology, Department of Electric Energy, O. S. Bragstads plass 2E, Trondheim 7034, Norway
d
University of York, School of Physics, Engineering and Technology, Heslington, York YO10 5DD, United Kingdom

A R T I C L E I N F O A B S T R A C T

Keywords: This review offers an in-depth examination of Deep Learning (DL) and Machine Learning (ML) techniques for
Deep learning smart grid load forecasting, emphasizing language precision, methodological rigor, and the exploration of novel
Ensemble methods contributions. The language used in this review is both technical and accessible, balancing complex concepts
Load forecasting
with clear explanations to cater to both specialists and general readers. It meticulously dissects contemporary DL
Machine learning
Smart energy management
models, including neural networks and ensemble methods, and evaluates their effectiveness through a detailed
Smart grid review of algorithms and frameworks. The methodology section systematically compares these techniques
against traditional forecasting methods using performance metrics such as MAPE, RMSE, and MSE, ensuring a
comprehensive assessment of their accuracy and scalability. A significant contribution of this review is its ex­
amination of real-world applications and case studies, which demonstrate how ML and DL techniques address
practical challenges in energy management, such as grid stability and demand forecasting. Furthermore, the
review introduces novel perspectives on the integration of probabilistic forecasting and ensemble methods,
which offer innovative approaches for managing energy demand uncertainties. By identifying current limitations
and proposing future research directions, this review not only advances the understanding of DL and ML ap­
plications in smart grids but also provides a foundation for future developments in this evolving field.

1. Introduction manifesting as voltage fluctuations and, in severe cases, blackouts

(Ranjan et al., 2021a; Latif et al., 2021; Anonymous, 2021a). Further­
Instances of highly variable distributed generation sources (Hussain more, disturbances in the network are conceivable (Kamrul et al., 2024).
et al., 2020a), such as electric vehicles (Ustun et al., 2012; Distributed Potential network disruptions are a concern, which could lead to
Energy Resources DER, 2011), photovoltaic systems (Distributed Energy significant complications (Ustun et al., 2021a). The implementation of
Resources DER, 2011; Hussain et al., 2020a), wind turbines (Dey et al., energy management systems allows achieving dual objectives: reducing
2020; Patil et al., 2022), and energy storage devices (Das et al., 2022a; peak loads during unforeseen periods and improving the equilibrium
Haq et al., 2023), present challenges to the stability of power and dis­ between supply and demand effectively (Hussain et al., 2022; Srivastava
tribution networks (Nadeem et al., 2019). These sources contribute to et al., 2022; Das et al., 2022b; Ustun and Hussain, 2020). This goal is
distributed generation, and their integration may potentially jeopardize accomplished through the establishment of energy management sys­
system stability (Latif et al., 2020; Hussain et al., 2020b; Barik et al., tems. Two fundamental categories can be applied to energy manage­
2021). The primary concern in many cases is the imbalance that can ment classification (Fekri et al., 2023). The first category pertains to the
arise between power supply and demand (Farooq and Rahman, 2022; supplier side, such as electric utilities, which involves the activation or
Safiullah et al., 2022). Such imbalances, observed in numerous case deactivation of certain generators in response to variations in load de­
studies, can lead to disruptions in the network (Aftab et al., 2021), mand (Ustun et al., 2022).

* Corresponding author.
** Corresponding author at: Norwegian University of Science and Technology, Department of Electric Energy, O. S. Bragstads plass 2E, Trondheim 7034, Norway.
E-mail addresses: [email protected] (B. Biswal), [email protected] (S. Deb), [email protected] (S. Datta), [email protected]
(T.S. Ustun), [email protected] (U. Cali).

https://doi.org/10.1016/j.egyr.2024.09.056
Received 3 June 2024; Received in revised form 19 September 2024; Accepted 23 September 2024
Available online 27 September 2024
2352-4847/© 2024 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

The second category, termed demand-side management, focuses on calendar effects. Methods such as ARIMA, exponential smoothing, and
consumer-side management of the market (Chauhan et al., 2021; Ranjan STLF applications frequently use machine learning techniques like
et al., 2021b). This entails consumers regulating their energy usage to gradient boosting and random forests.
meet the electricity demand created by generation-side implementation Medium-Term Load Forecasting (MTLF): MTLF typically spans from
of demand-side management. This is done to meet the power demand. several months to a year ahead. It facilitates capacity planning, infra­
The primary objective of implementing energy management is to reduce structure investment decisions, and policy formulation. MTLF models
operating and consumption costs (Singh et al., 2021), minimize energy incorporate factors such as economic indicators, population growth, and
losses (Sahoo et al., 2023). Another objective is to improve network industrial activities. Time series analysis, econometric models, and
reliability (Al-Shetwi et al., 2022; Dawn et al., 2021). regression techniques are commonly used in MTLF.
Load forecasting, illustrated in Figure 4, is resilient due to its capa­ Long-Term Load Forecasting (LTLF): LTLF involves predicting load
bility to adjust to dynamic changes in energy demand and generation demand several years or decades into the future. It informs long-term
patterns, thereby enhancing grid stability and reliability. By integrating investment strategies, grid expansion plans, and renewable energy
machine learning and deep learning techniques, forecast models can integration. LTLF models consider factors like demographic trends,
accurately capture intricate trends and fluctuations. The resilience is technological advancements, and regulatory changes. Econometric
further bolstered by ensemble methods, which mitigate uncertainties models, scenario-based analysis, and system dynamics approaches are
and enhance forecast performance. Additionally, probabilistic fore­ employed in LTLF due to the complexity and uncertainty of long-term
casting provides a proactive approach to managing unpredictability, forecasts.
empowering decision-makers to make informed risk management de­ In conclusion, each type of load forecasting has a specific function
cisions. Overall, the resilience of load forecasting ensures efficient and makes use of specialized techniques to deal with the particular
resource allocation, grid optimization, and sustainable energy man­ difficulties posed by various time horizons for prediction in power sys­
agement, thereby contributing to a more robust and adaptable energy tem planning and operation. On the other hand, the taxonomy for load
infrastructure. forecasting approaches can be organized into three primary categories:
The power utilities with different generation modalities (DGM) statistical methods, machine learning methods, and deep learning
experience complexities & minimal error in predicting future load methods. Here is a concise summary:
forecasting. It features diverse components including residential homes, Statistical methods play a crucial role in load forecasting and anal­
commercial offices, and industrial facilities, all connected to a central ysis. Time Series Analysis is employed for load forecasting, utilizing
power distribution tower. This highlights the process of collecting data strategies such as Exponential Smoothing and ARIMA (Auto Regressive
through smart meters installed at each component location. This data is Integrated Moving Average) (Dey et al., 2023). These techniques are
then transmitted to a central system for aggregation and analysis. The effective in capturing temporal patterns and trends in historical load
core of the process involves using advanced algorithms to forecast en­ data. Regression Analysis is utilized to establish relationships between
ergy demand and supply needs based on historical and real-time data. By historical load data and other significant variables (Hamoudi et al.,
optimizing electricity distribution, smart grids enhance energy man­ 2023). This method helps quantify the impact of various factors on
agement, improve efficiency, and balance supply and demand through electricity consumption patterns. Seasonal decomposition methods,
data-driven insights. such as the seasonal decomposition of time series, are employed to
On the other hand, it has a significant future in which the majority of decompose load data into trend, seasonal, and residual components
the research that is being done right now is concentrated on the devel­ (Yadav and Malik, 2021). This approach aids in understanding seasonal
opment of complex algorithms and models in order to improve the variations and their influence on overall load patterns.
management of the energy that is on the grid. This is being done in order Machine Learning methods offer advanced techniques for load
to make the grid run more efficiently. The reason for this is that it has a forecasting. Support Vector Machines (SVM) are utilized to predict
good future ahead of it. For the purpose of satisfying the requirements of future load by analyzing historical data and relevant characteristics
the customers and making their lives easier, it is essential that there be (Chatterjee et al., 2024). SVMs excel in capturing complex relationships
an increase in the quantity of power that is generated. The reason for this and patterns in data. Techniques such as Random Forest, an ensemble
is that the need for energy is only going to continue to increase all across method, is effective in handling non-linear relationships and in­
the world. On the other hand, the demand for electricity may result in teractions between variables in load forecasting (Ustun et al., 2021a,
difficulties for the operators of the electric utilities and the system as a 2021b; Yarar et al., 2023). By aggregating predictions from multiple
whole. This predicament is the result of a variety of variables, the most decision trees, Random Forest enhances predictive accuracy. Similar
important of which are the unpredictability of the electric load and the approaches iteratively build a series of weak learners to create a robust
fact that there are more customers overall. In addition, there is a learner for load forecasting tasks (Pattanaik et al., 2024). This iterative
considerable possibility that high peak demands may occur at a variety approach sequentially improves prediction accuracy by focusing on
of different times, which may pose a risk to the system’s capacity to previously mis predicted data points.
carry out its functions (Al-Quraan et al., 2023; Alsirhani et al., 2023). Deep Learning methods, particularly Recurrent Neural Networks
For effective energy planning and management in power systems, (RNNs) such as Long Short-Term Memory (LSTM) networks, are
load forecasting is crucial. It is divided into a number of groups ac­ increasingly employed for sophisticated load forecasting. RNN archi­
cording to the time prediction horizon. tectures, notably LSTM networks, are adept at analyzing sequential data
Very Short-Term Load Forecasting (VSTLF): This category typically like time series, which is crucial for accurate load forecasting
involves predicting load demand over minutes to a few hours ahead. (Abdolrasol et al., 2023; Anonymous, 2021b; Ulutas et al., 2020). These
VSTLF is crucial for real-time operations, such as unit commitment, networks excel in capturing long-term dependencies and temporal dy­
dispatch scheduling, and grid stability. Techniques like autoregressive namics in data, making them well-suited for modeling electricity con­
integrated moving average (ARIMA), artificial neural networks (ANN), sumption patterns over time.
and support vector machines (SVM) are commonly employed for VSTLF Each of these methods possesses distinct advantages and disadvan­
due to their ability to capture short-term dependencies and rapid load tages, and Depending on the specific characteristics of the load data and
fluctuations. forecasting requirements, their effectiveness may vary. Finding the best
Short-Term Load Forecasting (STLF): STLF extends the prediction strategy for a given forecasting task requires conducting tests and vali­
horizon to a day, a week, or up to a month ahead. It aids in optimizing dating findings. Based on the above discussions, this review provides a
resource allocation, energy trading, and economic dispatch. STLF comprehensive overview of recent advancements in smart grid load
models often integrate weather forecasts, historical load data, and forecasting methods. It underscores the critical importance of load

3655
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

forecasting within the realm of smart energy management, emphasizing load forecasting enables efficient decision-making at different adminis­
its pivotal role in optimizing energy distribution and consumption trative levels and aids in coordinating load balancing tactics throughout
strategies. The review explores the integration of advanced techniques the grid.
from deep learning and machine learning, demonstrating their efficacy Probabilistic Load Forecasting: Probabilistic load forecasting extends
in enhancing the accuracy and reliability of load forecasting models. beyond making single predictions by offering probabilistic forecasts that
Furthermore, it identifies and discusses significant challenges encoun­ encompass the inherent uncertainty in load forecasting. Probabilistic
tered in smart grid load forecasting, including issues related to data forecasting involves generating a probability distribution of future load
quality, scalability, and model interpretability. These insights lay the scenarios, allowing decision-makers to evaluate the probability of
foundation for future research directions aimed at addressing these various events and make well-informed risk management choices, rather
challenges and advancing the field. Additionally, the review evaluates than anticipating a single value. Probabilistic load forecasting often
the potential of ensemble methods in bolstering load forecasting per­ employs techniques such as ensemble methods, Bayesian methodolo­
formance, highlighting their capability to mitigate uncertainties and gies, and Monte Carlo simulations. This method is especially beneficial
elevate forecast accuracy in dynamic energy environments. for utility companies and grid operators to accurately measure and
control the unpredictability linked to fluctuations in power demand,
2. Review of load forecasting in power systems market conditions, and the integration of renewable sources.
By integrating these classifications into the load forecasting paper, a
The load forecasting problem in conventional and smart grids varies thorough examination of the many aspects of forecasting methods and
substantially due to the distinct characteristics and capacities of each their practical ramifications in the energy industry may be achieved.
grid system. Traditional systems commonly employ rudimentary sta­
tistical models and historical data to forecast future power demand in 2.1. Background
load forecasting. These models frequently fail to consider the ever-
changing nature of energy consumption patterns and have limited ca­ Improving the transmission and consumption of electrical power in a
pacity to adjust to changing circumstances. smart grid for smart energy administration requires the use of load
In addition, traditional grids do not possess the capacity to see and forecasting. Given the increasing demand for energy, In order to forecast
act in real-time, which creates difficulties in administering demand- load accurately and efficiently, Deep learning and machine learning
response programmes and incorporating renewable energy sources. must be applied. ML and DL algorithms utilize historical consumption
Load forecasting in smart grids use sophisticated technologies like sen­ patterns, weather conditions, and other pertinent data to forecast future
sors, smart metres, and communication networks to collect up-to-date electricity demand with greater accuracy compared to conventional
information on electricity consumption, generation, and system condi­ approaches.
tions. Smart grid load forecasting utilises advanced machine learning The Smart Grid increases the efficiency and dependability of elec­
and deep learning algorithms to analyse extensive data and detect tricity distribution by utilizing modern technologies, with load fore­
intricate trends in energy consumption. This allows for more precise and casting playing a crucial role in this system. Complex correlations within
detailed forecasts of electricity consumption, which helps improve the the data are captured by deep learning, or DL, models like neural net­
administration and optimisation of the power grid. works and machine learning (ML) models like regression and decision
Furthermore, smart grids facilitate the implementation of dynamic trees. enabling more detailed predictions. This proactive strategy helps
pricing mechanisms and demand-response programmes, which enable utility companies optimize resource allocation, minimize grid conges­
utilities to motivate users to modify their electricity consumption in tion, and avert future failures. Additionally, it gives customers the
accordance with real-time system circumstances. In smart grids, load knowledge they need to make educated decisions about how much en­
forecasting is characterised by its dynamic, adaptive, and data-driven ergy they use, which promotes a more economical and sustainable en­
nature, which sets it apart from conventional grids. This allows for ergy environment. When load forecasting is combined In the framework
enhanced efficiency, dependability, and sustainability in energy of a smart grid, machine learning (ML) and deep learning (DL) opera­
management. tional efficiency is increased and a more intelligent, resilient energy
When it comes to load forecasting, the classification into distribution infrastructure that can meet the ever-changing needs of modern society
of load forecasting, hierarchical load forecasting, and probabilistic load is developed (Zafar et al., 2023; Onteru and Vuddanti, 2023).
forecasting provides a systematic framework to tackle various parts of
the forecasting process. Below is a detailed analysis of each category: 2.2. Motivation
Forecasting the Distribution of Power Demand is the process of
estimating the amount of power is known as spatial load forecasting. It is The motivation behind this review is to address the pressing need for
needed at different points within a specific region or network. This effective load forecasting in Smart Grids for Smart Energy Management.
technique recognizes the regional variability in load patterns caused by With the increasing complexity of energy systems and the rapid inte­
factors such as population density, industrial activities, and climate gration of renewable energy sources, there is a critical demand for ac­
fluctuations. Methods such as spatial interpolation, geographic infor­ curate prediction models. By focusing on the integration of Machine
mation systems (GIS), and clustering techniques can be used to capture Learning techniques, particularly Deep Learning, this review aims to
spatial relationships and forecast the load at specific areas. Efficiently explore recent advancements and identify challenges in order to guide
managing generation, transmission, and distribution resources across future research efforts. Ultimately, the goal is to contribute to the
multiple regions is of utmost importance for utilities, making spatial development of transparent, resilient, and human-centric smart energy
load forecasting a critical task. management systems to meet the evolving needs of the energy industry.
Hierarchical Load Forecasting refers to the process of predicting the
electricity demand at different levels of a hierarchical structure, such as 2.3. Objectives of review
at the national, regional, or local level. Hierarchical load forecasting
acknowledges the hierarchical arrangement of the electricity grid, The review has multifaceted objectives designed to advance under­
encompassing distinct tiers such as national, regional, and local grids. standing and application in the field. Primarily, its goal is to assess the
This methodology entails predicting the demand at every level of the state of Deep Learning approaches today as they relate to load fore­
hierarchy, considering the combination and separation of load data. casting in smart grids. This entails a thorough examination of current
Forecasts can be created for many levels of electricity demand, including models. This involves a comprehensive analysis of existing models, al­
national, regional, and specific load zones within regions. Hierarchical gorithms, and frameworks to assess their effectiveness, accuracy, and

3656
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

scalability. adjust to fluctuating circumstances renders it very suitable for the dy­
Secondly, the review seeks to identify the practical implications of namic aspects of energy usage. The development of sophisticated models
implementing ML and DL for load forecasting in Smart Energy Man­ like This era saw major advancements in the estimation and prediction
agement. This includes evaluating their impact on grid reliability, en­ of energy usage over time produced by long short-term memory net­
ergy efficiency, and overall system performance. By delivering into real- works (LSTMs) and recurrent neural networks (RNNs) (Yao et al., 2022;
world applications and case studies, the objective is to distill valuable Islam et al., 2022; Khan et al., 2022; You, 2022).
insights that can inform future deployments and improvements. There was also a shift in the historical path toward real-time pre­
Lastly, the review aims to pinpoint challenges and limitations asso­ diction. Modern sensor technology and the widespread use of smart
ciated with ML and DL approaches, providing a critical examination of meters have made it possible for utilities to collect data more often,
potential hurdles in deployment and suggesting avenues for further which allows them to produce projections that are more accurate and
research. Through these objectives, the review strives to contribute to timely. Since renewable energy sources and electric vehicles have
the ongoing evolution of Smart Grids for smarter and more sustainable become more prevalent, the real-time functionality has become crucial
energy management. for controlling the increasing volatility & unpredictability of energy
consumption patterns (Complementary and Medicine, 2023; Torres
2.4. Scope and limitations et al., 2023; Wang et al., 2023a; Yang et al., 2023).
As they evolved, smart grids were employed for purposes other than
This review comprehensively examines the latest research trends and load forecasting. In order to maximize demand responsiveness, identify
achievements in enhancing smart grid load prediction accuracy using issues, and forecast maintenance needs, deep learning (DL) and machine
deep learning and machine learning techniques. It covers diverse learning (ML) techniques were applied. The goal of these all-inclusive
methodologies such as ensemble methods, time series analysis, and approaches is to create energy ecosystems that are more robust and
optimization techniques, highlighting the dynamic and evolving nature adaptive. Notwithstanding these developments, problems remained.
of this field. Emphasis is placed on key performance metrics like MAPE, The effectiveness of machine learning (ML) and deep learning (DL)
RMSE, RRSE, and MSE. The focus is on demonstrating the effectiveness models is significantly impacted by the quality and accessibility of data,
of complex algorithms, particularly neural networks, in addressing en­ according to historical trends. Biases or inaccuracies in prior data may
ergy system challenges, supported by empirical case studies that connect affect the prediction’s accuracy. Furthermore, the understandability of
theoretical advancements with practical applications. complex deep learning models has become a source for concern because
Despite promising advancements, the review identifies several lim­ awareness of the decision-making process is essential to promoting
itations, including the opacity of complex models and persistent cyber­ acceptance and confidence in real applications.
security concerns. These challenges highlight the need for ongoing In the future, load forecasting will continue to grow as learning with
research to develop transparent, resilient, and human-centric smart deep neural networks (DL) and ML technologies grow (Nagarajan et al.,
energy management systems. The review also underscores the impor­ 2022; Refaai et al., 2022). The historical review provides the foundation
tance of interdisciplinary cooperation to address emerging issues and for this progression. The incorporation of explainable AI, reinforcement
advance towards a more intelligent and sustainable energy future. learning, and hybrid models that leverage the advantages of different
Continuous exploration is essential to fully harness the potential of techniques can enhance the precision and practicality of load forecasting
advanced techniques in smart grids, ensuring their real-world applica­ in Smart Grids, ultimately leading to a more intelligent, efficient, and
bility and long-term efficacy. sustainable energy management approach.

3. Literature review 3.1. Comparative evaluation of author contributions in comparison to

other review papers
A revolutionary trend marked by advancements in technology and a
growing emphasis on sustainability can be seen in the historical devel­ When assessing the authors’ addition to the current collection of
opment of load forecasting utilizing machine learning and Deep learning review articles on smart grid load forecasting, it is crucial to examine the
methods for intelligent energy management in the framework of smart thoroughness of their study, the originality of their findings, and their
grids. incorporation of recent breakthroughs in machine learning (ML) and
Initially, load forecasting heavily depended on conventional statis­ deep learning (DL) techniques. The authors’ contribution is notable in
tical techniques, time-series analysis, and basic forecasting models. multiple ways when compared to prior review publications in the field.
These methods have constraints in their capacity to comprehend the Firstly, their historical overview offers a thorough analysis of the
intricate and ever-changing characteristics of energy usage patterns, development of load forecasting techniques, documenting the shift from
particularly in light of the growing urbanization and industrialization. conventional statistical methods to the incorporation of machine
With the increase in energy needs, it became clear that more advanced learning (ML) and deep learning (DL) approaches. The thorough his­
technologies were needed to tackle the changing issues of the power grid torical backdrop allows readers to understand the importance of recent
(Das et al., 2023; Kermani et al., 2023; Xin et al., 2022). achievements in the wider context of smart grid management.
The advent of machine learning signified a significant shift in the Particular uses of machine learning (ML) and deep learning (DL) in
existing methods. Machine learning algorithms, such decision trees and load forecasting, emphasising the efficacy of neural networks like long
regression analysis, started to become more and more common in load short-term memory (LSTM) and recurrent neural networks (RNNs) in
forecasting in the late 20th century. These methods provided enhanced collecting intricate patterns in energy consumption data is explored. The
precision by taking into account past consumption data, weather trends, authors provide useful insights into the potential of deep learning ap­
and other external variables. Machine learning introduced a methodol­ proaches to tackle the issues posed by dynamic energy usage patterns in
ogy that relies on data analysis to predict future outcomes, allowing smart grids, by emphasising on this advanced methodology.
utilities to make better-informed choices regarding the allocation of The authors highlight the significance of having the ability to predict
resources & the grid’s management. in real-time, especially in the face of growing instability and unpre­
The application of deep learning (DL) in the twenty-first century dictability resulting from the incorporation of renewable energy sources
significantly changed how load forecasting works inside smart grids. and electric vehicles. This perspective that looks ahead emphasises the
Deep learning, especially neural networks, demonstrated an impressive importance of their review in tackling the current and future issues that
ability to discover complex patterns and relationships inside large the energy business is facing. The authors of this review study also
datasets. DL’s capacity to autonomously extract characteristics and recognise the limitations and difficulties that come with ML and DL

3657
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

models, including problems with data quality and the absence of inter­ Table 2
pretability in intricate DL architectures. Comparison of Machine Learning (ML) and Deep Learning (DL) Models Based on
Through a thorough evaluation of these obstacles, the authors pre­ Validation Indices.
sent an impartial viewpoint on the tangible consequences of incorpo­ Author Methodology used Validation Indices Ref.
rating machine learning (ML) and deep learning (DL) methods in actual name/year
smart grid settings. Also suggest that future research should focus on Guo/2021 Deep neural network Simulation findings (Guo et al.,
integrating explainable AI and hybrid modelling techniques to improve (DNN)(Deep learning) demonstrate that the 2021)
the accuracy and applicability of load forecasting. The authors’ prediction model’s
performance indices’
commitment to moving the field beyond present restrictions is evident in
mean absolute
their forward-thinking approach. percentage error
This approach highlights the potential significance of their work on eMAPE and root-mean-
defining future improvements in smart grid management. The authors’ square error eRMSE are
contribution to the current literature on smart grid load forecasting is 10.01 % and
2.156 MW.
notable for its thorough analysis, integration of recent breakthroughs, .
and forward-thinking approach to future research paths. The authors’ Asiri /2024 Deep learning, LSTM, LFS-HDLBWO (Asiri et al.,
comprehensive analysis of machine learning (ML) and deep learning CNN, optimization, outperforms other DL 2024)
(DL) approaches in the context of smart grids offers useful insights that performance. methods with error
rates of 3.43 and 2.26.
contribute to the continuing discussion on sustainable energy manage­
Alquthami/ Short-term load 99.21 % accuracy in (Alquthami
ment. Tables 1,2 2022 forecasting using training, 100 % F1, et al., 2022)
machine learning 100 % precision,
3.2. Key concepts and definitions methods and improved 99.9 % recall, and
decision trees 99.70 % correctness in
. testing.
The main ideas and terms surrounding load forecasting in relation to
Muzumdar machine learning MAPE=1.18 (Muzumdar
the supply of electricity. For the electricity system to remain reliable and /2020 techniques include et al., 2020)
to avoid possible disturbances like blackouts, load forecasting accuracy Random Forest and
is essential. To gauge how accurate load forecasting methods are, a Support Vector
Regressor (SVR) (RF).
number of Key Performance Indicators (KPIs) have been identified.
These are the definitions and main ideas that were discussed. Statisti­
cally speaking, the Mean Absolute Error (MAE) measures the average
n
absolute deviations between the projected and actual values in a dataset. 1∑
MAE = |xi − x| (1)
A straightforward method to assess prediction accuracy is to use the n i=1
mean of absolute differences (MAE), which does not take error direction
into account. The calculation of the Mean Absolute Percentage Error (MAPE) in­
volves dividing the sum of all individual absolute mistakes by the total of
all real values. This statistic normalizes the forecast accuracy in respect

Table 1
Literature summary.
Author name /Ref Methodology used Gap/ problem definition Dataset used Parameter measured

Jiang (Jiang et al., Hybrid forecast-optimize tasks and Handling uncertainties in CityLearn Challenge 2022 dataset for Energy dispatch accuracy, robustness,
2023) efficient online data augmentation renewable energy integration and smart building energy management. adaptation to real-time data
scheme. real-time energy dispatch. distribution.
Chen (Chen et al., ARIMA and Bi-LSTM models for Accurate forecasting of solar One year of real-time solar power Mean Absolute Error (MAE) and Root
2023) solar power production prediction. power production for efficient production data for prediction. Mean Squared Error (RMSE).
grid management.
Maghraoui ( SVM, ANN, DT, and RF for hotel Optimizing energy efficiency in Hotel building energy consumption Accuracy of energy consumption
Maghraoui et al., energy consumption prediction. hotels to prevent grid overload. data from a case study in Shanghai. predictions using MAE, RMSE, and
2022) efficiency.
Alsharekh ( R-CNN and ML-LSTM for short- Accurate multistep electricity load Residential IHEPC and commercial Forecasting accuracy measured by
Alsharekh et al., term electricity load forecasting forecasting for smart grid PJM datasets for electricity load error rates compared to baseline
2022) framework. management. forecasting. models.
Teekaraman Optimizing smart grid video Optimizing energy consumption Wireless video sensor network Power, energy consumption, delay,
Teekaraman sensor networks for energy and quality of service in smart grid dataset for optimizing energy and transmission rate, delivery rate,
et al., (2022a) efficiency and performance networks. quality. convergence rate, quality.
Ibrar Predicting decentralized power Addressing imbalances and Data from an open machine learning Grid stability parameters: electricity
Ibrar et al., (2022) grid stability using machine enhancing prediction accuracy in library used to simulate a volume (p), cost-sensitivity (g),
learning and resampling. decentralized power grid stability. decentralized grid response times (tau).
.
Zhu DASG protocol using Chinese resolving issues with data Smart grid dataset, privacy- Smart grid parameters: status, privacy,
Zhu et al., (2022) Remainder theorem for smart grid integrity in smart grid aggregation preserving, integrity, DASG protocol, integrity, DASG protocol, encryption,
security. protocols. homomorphic encryption, aggregation.
aggregation.
Sodagudi (Sodagudi Hybrid control system Voltage quality, harmonic Dataset on power electronics, energy Battery SOC, charging voltage, battery
et al., 2022) methodology for efficient power distortion, and efficiency efficiency, and control systems current, ultracapacitor current, and DC
electronic interface in renewable challenges in power electronics. evaluation load current.
energy
Yu (Yu et al., 2022) 3-tier cloud-fog-consumer Network congestion and Dataset on cloud data center size, Measured parameters: Cloud size, VM
architecture; real-time VM imbalance in cloud data centers VM growth, and load balancing. growth, network resources, load
migration for load balancing. for VMs. balancing, response time, cost
optimization.

3658
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

to the scale of the data and presents it as a percentage. the comparative analysis of load forecasting and prediction modelling,
both machine learning (ML) and deep learning (DL) show promise. Both
1 ∑ |et |
MAPE = (2) strategies use statistical methods and algorithms to examine data trends
n |at |
and generate precise forecasts.
There are two steps involved in calculating the Root Mean Square Additionally, both ML and DL methodologies aim to optimize
Error (RMSE). To obtain the Mean Squared Error (MSE), the squared resource allocation and improve grid stability in energy management
errors are first averaged. After that, the square root of the MSE is used to systems.The analysis of ML and DL lacks specificity regarding their
calculate the RMSE, a gauge of forecast accuracy (Botchkarev, 2019). contributions to energy management load forecasting research. Future
n
directions for enhancing forecasting accuracy and addressing industry-
1∑ ̂ i )2 specific challenges are needed to provide a more targeted and insight­
MSE = (Yi − Y (3)
n i=1 ful perspective.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ While literature acknowledges challenges in energy management
√∑
√n forecasting such as data quality, model opacity, scalability, and dynamic
√ (Pi − Oi )2
√i=1 energy systems, gaps persist in providing comprehensive solutions.
RSME = (4) While some research touches on these challenges, practical imple­
n
mentation and validation in real-world scenarios are lacking. Future
According to its definition, the Root Relative Squared Error The ratio efforts should focus on developing scalable, interpretable models that
of total squared mistakes between predicted and actual values to total effectively address the complexities of dynamic energy systems,
squared errors between average value and actual values is known as the bridging the divide between theoretical advancements and practical
root mean square error (RRSE). It evaluates predicting effectiveness by applicability in energy management forecasting.
comparing it to a reference value.
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
√∑ 3.3. Main limitations and future research directions
√ (ft − at )2
RRSE = √ ∑ 2
(5)
(at − a) Existing works on electricity load forecasting face several limita­
tions, including issues with model accuracy, adaptability to dynamic
By comparing the standard deviation of anticipated mistakes to the environments, and scalability. Many traditional models struggle with
mean of actual data, the Coefficient of Variation (CV) measures the high error rates in complex scenarios and lack the flexibility to accom­
accuracy of predicting. An alternative way to define it is as the RMSE, or modate real-time changes in energy consumption. Additionally, current
Root Mean Square Error divided by the mean of real data, providing a approaches often fail to integrate diverse data sources effectively and
normalized accuracy evaluation. may overlook the impact of emerging technologies and user behaviours.
RMSE Future research should focus on developing more robust and adaptive
CV = (6) models that leverage advanced machine learning techniques, such as
a
hybrid models combining deep learning and traditional methods.
These measurements are essential for evaluating the effectiveness of
Exploring real-time data integration, incorporating user behaviour an­
load forecasting techniques and guaranteeing the stability of power
alytics, and addressing the challenges of scalability and generalization
supply networks. By quantifying forecast accuracy, they enable effective
will be crucial for advancing forecasting accuracy and improving smart
monitoring and optimization of energy production, minimizing costs
grid management strategies.
and system failures (Guo et al., 2021).
Based on their validation indices, the table offers a comparison of
3.4. Key results and comparison of existing techniques
various machine learning (ML) and deep learning (DL) models. Several
writers have commented on these indices across a range of years.
In recent literature, various techniques have been explored for
Guo (2021) utilised a deep neural network (DNN) in their prediction
electricity load forecasting. For instance, ARIMA models have been
model. The simulation results revealed a mean absolute percentage error
praised for their simplicity but are often criticized for inadequate
(eMAPE) of 10.01 % and a root-mean-square error (eRMSE) of
handling of non-stationary data, resulting in higher forecast errors. On
2.156 MW. Asiri (2024) investigated advanced deep learning methods
the other hand, Deep Learning models like Bi-LSTM have demonstrated
such as Convolutional Neural Networks (CNN) and Long Short-Term
superior accuracy, with lower MAE and RMSE compared to ARIMA, as
Memory (LSTM). They introduced a new model called LFS-HDLBWO,
seen in studies predicting solar power production. Additionally, R-CNN
which outperformed previous deep learning techniques. The LFS-
with ML-LSTM has shown significant improvements in multistep fore­
HDLBWO model achieved error rates of 3.43 and 2.26, demonstrating
casting tasks, outperforming traditional models by reducing error rates
its superior performance.
in both residential and commercial datasets. While Support Vector
Alquthami (2022) conducted a study that concentrated on machine
Machines (SVM) and Random Forest (RF) are effective for various en­
learning algorithms used for predicting short-term electricity use, with a
ergy prediction tasks, advanced deep learning models consistently
particular emphasis on improving decision trees. Their model demon­
deliver better performance in complex scenarios. This comparison un­
strated remarkable validation indices, including a recall of 99.9 %, an F1
derscores the shift towards integrating sophisticated algorithms and
score of 100 %, a precision of 100 %, a training accuracy of 99.21 %,
real-time data for improved forecasting accuracy.
and a testing accuracy of 99.70 %. Muzumdar (2020) utilised machine
learning techniques, specifically Support Vector Regressor (SVR) and
3.5. Challenges & gaps
Random Forest (RF), and achieved a Mean Absolute Percentage Error
(MAPE) of 1.18, demonstrating a high level of predicted precision.
The review identifies several challenges and gaps in the application
Every study demonstrates the potential and efficacy of diverse ap­
of DL and ML techniques for smart grid load forecasting. Key challenges
proaches in the domains of deep learning and machine learning for
include the need for large and high-quality datasets, the complexity of
distinct applications, such as load forecasting and prediction modelling.
model training, and the computational resources required. There are
Although DL models such as DNN, LSTM, and CNN perform well in
also gaps in integrating these advanced techniques with existing grid
certain situations, traditional ML techniques like SVR and RF also show
infrastructure, and in addressing issues related to data privacy and se­
comparable performance. This highlights the significance of choosing
curity. Furthermore, the review highlights a lack of standardized eval­
the right methodology depending on the individual job and dataset. In
uation metrics and benchmarks, which complicates the comparison of

3659
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

different models. Addressing these challenges and gaps is crucial for the 3.7. Significant milestones
broader adoption and effectiveness of DL and ML in smart grid
applications. The use of ML and DL in Smart Grid demand forecasting for Smart
The review’s methodology involved a systematic comparison of DL Energy Management has experienced numerous noteworthy achieve­
and ML techniques against traditional forecasting methods. Key per­ ments, demonstrating the progress and transformational changes in
formance metrics such as MAPE, RMSE, and MSE were used to evaluate energy forecasting methods and technologies. Integrating ML and DL
accuracy and scalability. The analysis began with a comprehensive algorithms into load forecasting models is a major achievement. This
literature review to identify state-of-the-art models and algorithms. departure from typical statistical methods allows the models to adapt to
These models were then applied to real-world datasets to assess their complex energy usage data patterns and non-linear linkages. RNNs, or
effectiveness in practical scenarios. Additionally, the review included a recurrent neural networks, and LSTMs have significantly better load
meta-analysis of case studies and real-world applications to illustrate the estimates, notably for temporal linkages and complicated patterns.
practical challenges and successes of implementing DL and ML in smart Intelligent meter live data streams are another major breakthrough.
grid load forecasting. This rigorous approach ensured a thorough and Smart meters gather data often and continuously, it makes machine
balanced evaluation of the techniques. learning (ML) and deep learning (DL) possible.
AI & ML models to adapt to shifting load patterns (Kumar et al.,
3.6. Evolution of research in the field 2023; Healthcare Engineering, 2023). Real-time information availabil­
ity improves forecast accuracy and makes it possible to put more flexible
There have been several significant developments in the field of load energy management techniques into practice.
forecasting research as it relates to smart grids and machine learning and Ensemble learning has greatly increased load forecasting model
deep learning for intelligent energy management. These advancements dependability. Ensemble techniques combine model forecasts for accu­
exemplify an ongoing pursuit for more precise, adaptable, and efficient racy and reliability. Bagging and boosting reduce uncertainty and
techniques for forecasting. At first, research mostly concentrated on improve load forecasting systems. Applying Explainable AI (XAI) tech­
conventional statistical techniques for load forecasting. Nevertheless, niques in load forecasting is a recent achievement that focuses on
with the emergence of machine intelligence, specifically in the context making ML and DL models more interpretable. As these models increase
of regression models and time series analysis, researchers initiated in­ in complexity, comprehending the reasoning behind their predictions
vestigations into more advanced methods for capturing intricate pat­ becomes essential. Integrating explainable Artificial Intelligence (XAI)
terns and interconnections in energy consumption data. The techniques facilitates the understanding of how models make decisions,
introduction of neural networks—more especially, neural networks with promoting transparency and building confidence in the accuracy of the
recurrent architectures and long short-term memory (LSTMs and forecasted results.
RNNs)—caused a significant change. These deep learning architectures The incorporation of edge computing in load forecasting is a devel­
have shown exceptional ability in managing temporal dependencies, oping achievement. Edge devices, such as sensors and smart meters,
allowing for more precise forecasts of load patterns, particularly in sit­ have the ability to do calculations locally because to their built-in pro­
uations with dynamic and nonlinear connections (Ma et al., 2021; Zhang cessing capabilities. By minimizing latency, the efficiency of load fore­
et al., 2020). casting models is enhanced, particularly in real-time applications (Zhou
Ensemble learning approaches have evolved as a crucial field of et al., 2020; Xu et al., 2020; Singh and Khan, 2017). Finally, The
study, utilizing the combined intelligence of numerous models to discipline has advanced toward more accurate, detailed forecasting of
improve the accuracy and reliability of predictions. Bagging and load in smart grids with the application of machine learning (ML) and
boosting approaches gained popularity as they provided effective tech­ deep learning (DL) techniques. Adaptable, and transparent energy
niques to reduce uncertainty and enhance the overall dependability of forecasting systems. The continuous development of smart energy
load projections. The utilization of real-time data obtained from smart management is driven by computational breakthroughs, real-time data
meters became a primary emphasis, allowing for ongoing monitoring integration, and the emergence of new technologies.
and prompt adjustment to fluctuating load situations.
This advancement was a notable achievement in improving the 4. Methodology
speed and precision of load forecasting models, especially under
constantly changing conditions. Explainable AI (XAI) strategies have The study’s selection criteria for Load forecasting in Smart Grid for
become popular in order to tackle the issue of interpretability in Smart Energy Management using ML DL employ a targeted approach to
complicated machine learning & deep learning models (Yang et al., ensure the incorporation of pertinent and high-caliber literature. In­
2020). As these models became more complex, comprehending the un­ clusive sources must explicitly center on load forecasting applications
derlying reasoning behind forecasts became essential for establishing within the utilizing deep learning (DL) and machine learning (ML)
trust and acceptance. The incorporation of Explainable Artificial Intel­ methods in a smart grid setting to provide intelligent energy manage­
ligence (XAI) methodologies allowed researchers to provide significant ment. Academic papers, conference proceedings, and esteemed journals
insights into the model’s decision-making processes, hence augmenting presenting empirical research, case studies, or inventive methodologies
the lucidity and comprehensibility of load forecasting results. at this specific intersection are deemed suitable. Only peer-reviewed
Edge computing has lately emerged as a new and innovative area of publications in English are considered to uphold scholarly rigor. Ex­
study in load forecasting (Wang et al., 2023b). Integrating processing clusions consist of non-peer-reviewed sources, promotional materials,
capabilities at the edge, in close proximity to the data source, reduces and studies lacking direct relevance to load forecasting in smart grids
delay and improves the effectiveness of forecasting models, especially in with ML and DL applications. These refined selection criteria aim to
real-time scenarios. To summarize, as load forecasting research in smart assemble a focused and credible collection of sources conducive to an
electricity systems using ML and DL has advanced, traditional statistical exhaustive exploration of the topic (Wawale et al., 2022; Bolla et al.,
techniques have given way to more advanced, flexible, and transparent 2022).
methodology. The sector is constantly evolving, with ongoing ad­ Improved prediction accuracy and more economical energy use can
vancements that tackle obstacles and open up new opportunities for result from the application of machine learning (ML) and deep learning
effective and dependable smart energy management. (DL) techniques to load forecast in smart grids for smart energy
management.
Thus, it is possible to use multiple data sources. History of energy use
data is a valuable source of knowledge since it sheds light on patterns

3660
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

and variances in use across time. Comprehending the historical context 4.1. Search strategy
is essential for training machine learning (ML) and deep learning (DL)
models to appropriately identify and react to seasonal fluctuations, dy­ The search technique entails a methodical and focused attempt to
namic shifts in energy consumption, and recurrent patterns. Smart meter obtain pertinent literature and research works in the topic. Commencing
real-time data is a vital and useful source of regular, in-depth updates on the exploration on scholarly resources such as IEEE Xplore, Science
energy consumption. Since smart meters provide a steady stream of Direct, and PubMed provides the opportunity to retrieve peer-reviewed
data, deep learning (DL) /machine learning (ML) algorithms can swiftly publications, conference papers, and journals pertaining to smart grid
adapt to erratic fluctuations in demand. technologies, load forecasting, and machine learning applications in
By making them more responsive, this improves the efficacy of load energy management. To ensure precision in search queries, it is advis­
forecasting algorithms (Almadhor et al., 2022). Meteorological data able to utilize terms such as "load forecasting," "smart grid," "smart en­
serves as an additional powerful source of information. The amount of ergy management," and "machine learning."
energy used is directly influenced by the weather, and models that take Aside from academic databases, conducting searches on online re­
weather-related variations in energy demand into consideration can be positories like as arXiv and Google Scholar expands the range of avail­
developed with the use of meteorological data. Weather factors like able resources to encompass preprints, technical reports, and
temperature, humidity, and energy consumption can be correlated with publications that might not be found in conventional databases. Utiliz­
one other using ML/DL techniques. ing boolean operators, such as and or, in conjunction with phrase
This approach has the potential to greatly increase the accuracy of searching enhances the precision of the search outcomes.
load forecasting in smart grids. Demographic and economic data provide Analyzing relevant workshop and conference data about machine
significant contextual information. Comprehending the demographic learning and smart grid applications, such as those from the IEEE PES
makeup of a region and its economic operations helps in forecasting General Meeting or The International Conference on Clean Energy
energy consumption patterns by considering elements such as popula­ Technologies and Smart Grids (ICSGCE), offers valuable insights into the
tion density, industrial activities, and economic trends. Moreover, most recent advancements and research patterns in this domain.
comprehensive grid infrastructure data, encompassing details about the Browsing through the institutional repositories of universities and
structure and capability of the electricity distribution network, is research organizations aids in locating theses, dissertations, and tech­
necessary. nical publications pertaining to load forecasting in smart grids
By taking into account the grid’s capabilities and limitations, this (Teekaraman et al., 2022b; Li et al., 2022; Boum et al., 2022; Anantha
data enables deep learning (DL) and machine learning (ML) models to Krishnan et al., 2022).
make sure that load estimates align with the network’s operating pa­ Government publications from energy agencies and organizations,
rameters (Nayagam et al., 2022; Sujatha et al., 2022; Zhang et al., 2022). such as the Department of Energy (DOE), can provide valuable infor­
A complete data collection technique is employed in smart grid load mation on policy aspects and practical applications. Moreover, the
forecasting, encompassing historical consumption data, real-time data analysis of references in important papers and reviews aids in the
from smart meters, meteorological conditions, demographic insights, identification of influential works and establishing connections with a
and grid architecture characteristics. Through the integration of several wider range of pertinent literature. For the most updated information on
data sources, machine learning (ML) and deep learning (DL) models may the most recent advancements in the subject, you should subscribe to
generate accurate and flexible load predictions, hence augmenting the alerts for relevant keywords and monitor recently published articles.
efficacy of smart energy management. Figs. 1–3 With regard to load forecasting in smart grids, this thorough search
approach seeks to gather a broad range of sources, including theoretical
frameworks, empirical research, and real-world applications. The
emphasis is on applying deep learning and machine learning techniques
to smart energy management.

4.2. Inclusion and exclusion criteria

To ensure the Caliber, applicability and focus of the data collected,

inclusion and exclusion have been developed. The inclusion criteria are
satisfied by academic magazines, conference procedures and publica­
tions that in particular deal with the load forecasts in the context of the
intelligent grids. To create intelligent energy management techniques,
deep learning (DL) and machine learning (ML) in these projects. Pre­
sentations should include examples of cases, new techniques or legiti­
mate research on the relationships between different fields. The
publications that do not discuss in particular the applications of machine
learning (ML) and the deep learning (DL) are subject to the exclusion
criteria. which are not directly related to load forecasting in smart grids.
To maintain consistency and cognitive integrity, advertising messages
(Ghislain et al., 2022) are used.
The analysis of database results includes several critical dimensions.
It starts with assessing the number of articles retrieved and the specific
search terms used and provides insight into the scope and relevance of
the search query. In addition, reviewing the database sources consulted
provides insight into the breadth and comprehensiveness of the litera­
ture reviewed. Analyzing the distribution of publications over time re­
veals trends in research activity and development in the field, shedding
light on the evolution of researchers’ interest. Metrics such as keyword
frequency and citation counts are used to assess the depth and impact of
Fig. 1. Flowchart of the Methodology used in this research. the literature retrieved from databases, thereby assessing the

3661
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

Fig. 2. CNN LSTM Model (Livieris et al., 2020).

research is required on advanced deep learning architectures like

recurrent neural networks (RNNs) and (LSTMs) networks.
An essential aspect pertains to the incorporation of up-to-the-minute
data derived from smart meters and other sensors. This section examines
the optimal utilization of continuous, high-frequency input streams to
enhance the load forecasting models’ responsiveness and accuracy,
especially in situations that are dynamic and changing quickly.
The primary objective of this branch of artificial intelligence is to
create models and algorithms that are capable of giving comprehensible
justifications for their choices and actions. The need of enhancing the
interpretability of deep learning (DL)/machine learning (ML) models
has increased due to their growing complexity. Gaining more insight
into the decision-making processes used by models is the aim of
Explainable Artificial Intelligence (XAI) research. This knowledge con­
tributes to increased openness and confidence in the precision of fore­
casted results (Su, 2022; Xiong et al., 2022; Prasath et al., 2022; Khan
Fig. 3. Random Forest and Gradient Boosting (Kowalek, 2019).
et al., 2021a, 2021b).
Ensemble learning delves into the amalgamation of many forecasting
importance and influence of the research findings. Together, these ele­ models to generate ensemble approaches. The effectiveness of bagging
ments form a holistic approach to analyzing database results, providing and boosting strategies is examined in order to reduce uncertainties,
valuable insights into the current state and dynamics of scientific enhance resilience, and ensure accurate load predictions.
research on the topic. A burgeoning field of study focuses on the incorporation of edge
Search refinement involves several systematic steps to improve ac­ computing into load forecasting. Scientists are investigating the poten­
curacy and comprehensiveness. Initially, the integration of synonyms tial benefits of processing data at the edge, which is closer to where the
and specific terms ensures that the search effectively captures the rele­ data is generated. This approach aims to reduce delays, increase effi­
vant literature. Studies are selected according to clear criteria based on ciency, and make real-time load forecasting applications more practical.
relevance, methodology and publication date. In addition, screening of Certain studies explore the amalgamation of load forecasting with
additional databases will help to fill possible gaps in the existing liter­ renewable energy sources. This means that demand forecasting models
ature. Advanced filters are applied to refine the results with a focus on must be improved to account for the erratic nature of renewable energy
quality and relevance. Thorough selection processes ensure that relevant sources and account for variations in energy production.
studies that meet the defined criteria are included. Research on load forecasting in Smart Grids is categorized along
Smart meters provide real-time data on electricity consumption and several dimensions: algorithmic development, temporal dependency
provide instantaneous information. Typically, data collection intervals management, real-time data integration, interpretability, ensemble
vary based on deployment and utility needs, ranging from every learning, and the combination of edge computing and renewable energy
15 minutes to hourly intervals in practice. sources. Each element has an impact on the overall development of
intelligent energy management systems (Alazzam et al., 2021; Meh­
5. Classification of studies mood et al., 2021; Srisomboon et al., 2021).

Research on load forecasting in smart grids for smart energy man­

5.1. Thematic grouping
agement can be divided into a number of significant areas that cover
diverse aspects of this multidisciplinary topic and make use of deep
The combination of machine learning (ML) and deep learning (DL)
learning and machine learning (ML and DL) (Hasan et al., 2022; Ganesan
techniques in smart grids has substantially changed energy manage­
et al., 2022; Balasubramaniam et al., 2022).
ment, especially in load forecasting. This cutting-edge technology makes
The creation and optimization of machine learning (ML) and deep
it easy to design incredibly complex and adaptable systems for opti­
learning (DL) algorithms, with a focus on load forecasting, have been the
mizing energy use.
subject of extensive research. In order to improve prediction accuracy,
"Smart Energy Management" is a primary area of emphasis in the­
this entails making advancements in neural network topologies, opti­
matic classification. Deep learning (DL) and machine learning (ML)
mization methods, and ensemble learning approaches.
models are crucial in forecasting energy usage patterns, enabling pro­
The efficient management of temporal dependencies in load fore­
active decision-making to optimize resource allocation. By analyzing
casting models is a task taken on by academics. In order to efficiently
historical data and taking into account several important factors, these
capture time series data’s sequential patterns and dependencies, more
models are crucial in accomplishing accurate load forecasting. Taking

3662
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

this proactive stance increases grid reliability and makes it easier to improved effectiveness, dependability, and environmental friendliness
successfully integrate renewable energy sources, which increases sus­ in smart grids. A notable accomplishment in this domain is the sub­
tainability (Duangsuwan et al., 2021; Zhong et al., 2021; Tang et al., stantial enhancement in load forecasting precision facilitated by deep
2021; Xu et al., 2021). learning (DL) and machine learning (ML) models. These advanced al­
The "Smart Grid" itself is a further focus. The smart grid infrastruc­ gorithms use previous consumption data, weather trends, and other
ture can adapt more flexibly to changing demand patterns thanks to pertinent elements to generate more accurate energy demand estimates.
machine learning/deep learning algorithms, which reduce energy loss The improved precision is essential for optimising grid operations,
and increase grid efficiency. The grid can now automatically adjust and guaranteeing a steady power supply, and facilitating efficient resource
react quickly to changes thanks to these technologies, making the energy allocation. The versatility of Smart Grids equipped with Machine
distribution system more resilient and responsive. Learning (ML) and Deep Learning (DL) models is emphasised by their
Additionally, load forecasting is made more sophisticated and so­ capability to handle variations in energy consumption patterns. These
phisticated by the incorporation of machine learning (ML) and deep systems have the ability to promptly react to fluctuations in demand by
learning (DL) techniques. This is known as "Intelligent Grid Operation." analysing and interpreting data in real-time, thus improving the dura­
With the use of sophisticated analytics, these technologies help utilities bility and adaptability of the energy distribution grid. The ability to
more effectively distribute electricity, predict periods of peak demand, adapt is crucial when it comes to incorporating renewable energy
and manage the power system. This intelligence enhances energy con­ sources, which by nature are subject to change, into the power grid.
servation, reduces expenses, and ensures a more dependable power The results further emphasise the significance of deep learning (DL)
supply for end users. and machine learning (ML) in promoting proactive energy management
Ultimately focused on the subjects of intelligent grid operation, measures. These technologies enable utilities to forecast periods of
smart energy management, and smart grid augmentation. These the­ highest demand, optimise the distribution of energy, and execute effi­
matic areas work together to propel the energy sector toward a more cient demand control approaches. This not only reduces energy wastage
intelligent, efficient, and sustainable future by fusing machine learning/ but also contributes to sustainability objectives by encouraging more
deep learning techniques. effective utilisation of resources. The utilisation of ensemble techniques,
which merge projections from numerous models, has greatly enhanced
5.2. Methodological approaches the dependability of predictions. Ensemble approaches enhance the
accuracy and reliability of future energy needs by reducing the influence
Many methodological strategies are employed, including deep of individual model biases and uncertainties. This approach showcases
learning (DL) and machine learning techniques. in load forecasting for the adaptability of deep learning (DL) and machine learning (ML) in
Smart Grids. The goals are precise forecasting and effective energy tackling the difficulties of load forecasting in ever-changing Smart Grid
management. One important methodological approach involves situations.
analyzing historical data. ML and DL models utilize historical con­ Another crucial finding is the imperative need for ongoing model
sumption patterns, weather conditions, and other pertinent information retraining. Ensuring that ML and DL models are regularly updated to
to detect trends and correlations. Through the analysis of these historical reflect changing energy usage patterns guarantees that these models
data sets, the models can acquire knowledge and adjust to the distinct maintain constant accuracy as time progresses, these improvements
attributes of energy consumption patterns, hence enhancing their ability underscore the significant influence of deep learning (DL) and machine
to generate more accurate forecasts. Another method involves feature learning (ML) technologies in determining the future of energy man­
engineering, which involves incorporating domain-specific information agement. The enhanced accuracy in load prediction and the adaptability
to improve the model’s comprehension of the data. Engineers and data of Smart Grids lead to a more intelligent, efficient, and environmentally-
scientists analyze crucial factors that influence energy usage, such as friendly energy future. The ongoing integration of machine learning
holidays, special events, and economic indicators. These features are (ML) and deep learning (DL) into energy management systems is
added to the model, which improves prediction accuracy and increases advancing. These findings provide vital insights into the crucial role
system flexibility in response to varying energy demand. these technologies will have in the development of the future Smart
Ensemble methods offer an alternative strategy by amalgamating Grids and their contribution to a robust and sustainable energy
predictions from numerous models to attain a more resilient and precise landscape.
forecast. Ensemble approaches boost the overall reliability of load
forecasting by combining results from various machine learning (ML) 6.1. Trends & patterns
and deep learning (DL) models, thereby reducing the risk of individual
model biases. The energy systems sector is undergoing a significant trans­
Furthermore, the ongoing process of updating and refining the model formation, driven by key trends and patterns in load forecasting for
is an essential methodological approach. ML and DL models are dynamic Smart Grids, particularly through the use of Machine Learning (ML) and
and require frequent modifications to accommodate changing patterns Deep Learning (DL). A prominent trend is the growing reliance on
and variables. Ongoing retraining guarantees that the models stay up-to- advanced ML and DL methods for load forecasting. These technologies
date and efficient in reflecting the complexities of evolving energy usage enable more accurate predictions by analyzing large datasets and ac­
patterns over time. The methodological approaches in load forecasting counting for various factors, allowing for better adaptation to dynamic
for Smart Grids utilizing ML (Machine Learning) and DL (Deep Learning) shifts in energy usage patterns. This shift represents a move away from
involve analyzing historical data, creating relevant features, employing traditional forecasting techniques toward data-driven, intelligent ap­
ensemble methods, and continuously retraining the models. These ap­ proaches, resulting in unprecedented accuracy in predicting future en­
proaches jointly enhance the creation of precise, flexible, and robust ergy needs. Another key trend is the increasing emphasis on real-time
energy management systems within the framework of Smart Grids. responsiveness within Smart Grids. ML and DL models empower Smart
Grids to continuously analyze incoming data and generate insights,
6. Key findings enabling swift reactions to changes in energy demand. Real-time
responsiveness is crucial for optimizing grid operations, managing
The combination of deep learning (DL) and machine learning (ML) renewable energy integration, and ensuring a robust and adaptable
technologies has made significant progress in energy management by energy distribution system that can meet the evolving demands of a
incorporating several study areas of load forecasting in smart grids. dynamic society. The growing popularity of ensemble approaches in
These technologies have introduced a new period characterised by load forecasting is another significant trend. By combining predictions

3663
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

from multiple models, these methods harness the strengths of different (ML) and deep learning (DL) models heavily depend on extensive
algorithms, reducing individual model limitations and enhancing the datasets for the purposes of training and validation. However, the
overall reliability of predictions. Ensemble methods are particularly presence of inconsistent, inadequate, or flawed data might hinder the
valuable in achieving more resilient and precise forecasts in the face of accuracy of their predictions. Smart Grid technology continues to face
uncertainty. Additionally, there is a clear movement towards integrating the difficulty of consistently obtaining high-quality data from several
Smart Energy Management principles into broader sustainability ini­ sources, the opaque nature of numerous deep learning (DL) and machine
tiatives. ML and DL techniques in load forecasting enable utilities to take learning (ML) models poses challenges in comprehending their decision-
proactive steps in managing energy resources, improving grid efficiency, making mechanisms. Transparency and interpretability are essential,
and promoting energy conservation. This aligns with global efforts to particularly when stakeholders require comprehension of the variables
transition to cleaner and more sustainable energy systems, marking a impacting energy estimates. Cybersecurity is a major issue of concern.
significant shift toward intelligent and environmentally conscious en­ Cyberattacks can exploit weaknesses that arise from the integration of
ergy management practices. These trends and patterns highlight a major Machine Learning (ML) and Deep Learning (DL) technologies in Smart
transition towards intelligent, adaptive, and sustainable energy systems. Grids. To guarantee the dependability and safety of Smart Grids, it is
The integration of advanced forecasting techniques, real-time adapt­ imperative to implement strong cybersecurity protocols in conjunction
ability, and ensemble methods points to a future where technology plays with the creation and upkeep of Machine Learning (ML) and Deep
a critical role in optimizing energy use, enhancing grid reliability, and Learning (DL) models. To tackle these difficulties, it is necessary to foster
fostering a robust and sustainable energy ecosystem. collaboration among specialists in energy systems, data science, cyber­
security, and regulatory frameworks. In order to fully harness the ca­
6.2. Emerging themes pabilities of load forecasting in Smart Grids and guarantee the secure
and efficient deployment of intelligent energy management systems, it is
One emerging theme is the pursuit of Explainable Artificial Intelli­ imperative to address these challenges as the industry progresses.
gence (XAI) in energy forecasting models. As ML and DL techniques
advance, there is an increasing recognition of the need for transparency 7.1. Deficiencies in current literature
and interpretability in these models’ decision-making processes. The use
of complex algorithms, particularly deep neural networks, often results The literature evaluation emphasises various significant research
in "black box" models, where the internal workings are difficult to un­ gaps that have been highlighted in recent publications. The gaps
derstand. This lack of transparency poses challenges in critical appli­ encompass uncertainties in managing energy in smart grids, the struc­
cations like energy management, where stakeholders need a clear tural aspects of constructing large-section shield tunnels, the utilisation
understanding of the reasoning behind specific forecasts or actions. XAI of machine learning in urban energy systems, the integration of
aims to address this issue by integrating transparency into ML algo­ renewable energy, and the recognition of cities using machine vision.
rithms. Practitioners are developing methods to interpret and under­ (Khan et al., 2022) explicitly identifies deficiencies in the manage­
stand the predictions made by these models. In load forecasting, XAI ment of energy in smart grids, highlighting the necessity for a more
seeks to clarify how the model incorporates historical data, accounts for thorough examination. These gaps include areas of uncertainty, strate­
external variables, and assigns importance to different factors when gies for optimisation, and the integration of stakeholders, all of which
making predictions. The growing importance of XAI is driven by its are essential for guaranteeing stability and efficiency in smart grid op­
ability to build trust and encourage broader adoption of advanced pre­ erations. This highlights the need for more extensive study to tackle
dictive models in the energy industry. Stakeholders such as utility pro­ these intricacies and improve the overall efficiency of smart grid energy
viders, regulatory bodies, and end-users often require a thorough management systems.
understanding of the factors influencing energy forecasts to make (You et al., 2022) identified a research gap related to the
informed decisions, allocate resources efficiently, and maintain grid load-bearing capacity and large-section shield tunnel failure character­
stability, the incorporation of XAI aligns with regulatory requirements istics at high water pressure.
that demand transparency in decision-making processes, especially in The call for comprehensive evaluation and control procedures
critical infrastructure like Smart Grids. It also addresses ethical concerns highlights the need for a deeper understanding of the structural aspects
by promoting transparency and reducing the risk of biases in decision- of such tunnels, pointing towards a gap in the existing literature on
making, thus supporting the responsible and reliable integration of AI tunnel engineering and safety under challenging conditions.
in energy management. (Almadhor et al., 2022) emphasized a dearth of research on machine
As the energy sector increasingly leverages ML and DL for load learning methods for calculating electrical power requirements in urban
forecasting, XAI is becoming a crucial consideration. XAI enhances settings and encouraging the production of renewable energy. This un­
transparency and interpretability, ensuring the reliability and accep­ derscores a gap in literature regarding the application of machine
tance of advanced forecasting models. Additionally, it provides a foun­ learning techniques to address energy challenges in urban environ­
dation for the ethical and responsible application of AI in the Smart Grid ments, where sustainability is a growing concern.
domain. (Nayagam et al., 2022) drew attention to the lack of grid integration
of hybrid renewable energy sources’ energy management systems. The
7. Key challenges & knowledge gaps disparity necessitates more investigation into accurate and effective
strategies and technology for incorporating renewable energy sources
Based on the observations of evolving trends and patterns, it is into the systems that exist today.
crucial to recognise that implementing load forecasting in Smart Grids (Liu and Liu, 2022) highlighted a research gap in machine
utilising Machine Learning (ML) and Deep Learning (DL) techniques is vision-based intelligent city recognition, particularly in maximizing the
not free from difficulties. The intricacy of the energy system poses a effectiveness and precision of threshold segmentation algorithms. This
significant obstacle. Smart Grid ecosystems consist of various inter­ indicates a need for more detailed investigations into the challenges and
connected elements, such as different energy sources, varying con­ advancements in machine vision applications for intelligent city recog­
sumption patterns, and changing external influences. To accurately nition, focusing on improving segmentation algorithms for enhanced
anticipate loads, machine learning (ML) and deep learning (DL) systems accuracy.
must navigate intricate linkages and adapt to altering conditions, To summarize, the gaps identified in the literature emphasize the
capturing the full spectrum of this complexity. The process is further necessity for further research in several areas, including smart grid en­
complicated by the accuracy and availability of data. Machine learning ergy management, tunnel engineering, machine learning applications in

3664
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

urban energy systems, integrating renewable energy sources, as well as 8.2. Areas for enhancement
machine vision-based intelligent city recognition. Rectifying these de­
ficiencies would enhance the advancement of more resilient and effec­ These problems present numerous opportunities for improving
tive systems within the realm of intelligent power grids and urban research and practical applications. Enhancing the interpretability of
sustainability. models presents a valuable chance to establish confidence and
comprehension among stakeholders by tackling the problem of opacity.
8. Methodological critique Furthermore, there is a possibility for advancement in data quality and
pre-processing techniques to tackle inconsistencies and biases. Investi­
The evaluation of research techniques highlights the dynamic and gating innovative ensemble methods and optimisation strategies could
diverse nature of load forecasting for Smart Grids, based on the previous significantly improve the resilience and effectiveness of load forecasting
discussion of trends and issues in this sector. Researchers utilise a range models. The collaboration between academia and industry can enhance
of methods to tackle the intricacies of load forecasting and intelligent the widespread implementation of research findings and the incorpo­
energy management, with Machine Learning (ML) and Deep Learning ration of practical knowledge. Examining energy storage alternatives
(DL) playing crucial roles. These methods entail the creation and and incorporating them into forecasting models offers a chance to
refinement of algorithms that utilise past consumption data, weather enhance the resilience and dependability of the power grid in the face of
patterns, and other pertinent aspects to produce accurate predictions. changing supply and demand trends.
Typical methods involve using supervised learning algorithms like
regression and neural networks, which can analyse intricate patterns in 9. Integration and synthesis
data to generate precise predictions. Time series analysis is a widely
used approach in load forecasting. This methodology carefully analyses Comparative analysis reveals that various studies employ distinct
past data to detect repetitive patterns and trends within certain time methodologies and provide diverse contributions. The repeated motif
periods. ARIMA and its variations are frequently employed to address revolves around the utilization of deep learning (DL) and machine
temporal dependencies in energy consumption patterns. learning methodologies, underscoring their importance in raising the
Ensemble approaches, which combine predictions from numerous accuracy of load forecasting. Certain studies highlight the use of
models, have become well-known for their efficacy. Methods such as ensemble techniques, including merging forecasts from several models,
bagging and boosting leverage the advantages of different algorithms, to improve load forecasting’s resilience. By using this method, the
leading to load estimates that are more resilient and precise. These shortcomings of individual models are lessened and the accuracy of
strategies aid in reducing the constraints of individual models, hence energy demand predictions is increased.
improving the overall reliability of predictions. Optimisation strategies By contrast, other studies highlight the variety of approaches in the
improve the efficiency of energy management systems. Metaheuristic field by concentrating on model parameter optimization through the use
techniques, such as genetic algorithms and particle swarm optimisation, of metaheuristic algorithms. In addition, a noteworthy trend is the in­
are used to optimise model parameters, enhancing their performance in clusion of real-world case studies. Research that employ real Smart Grid
load forecasting. data to validate their models offer useful information about the suit­
the assessment of these approaches strongly depends on case studies ability and efficiency of various forecasting techniques (Zhou et al.,
and practical implementations. Researchers verify the accuracy and 2022; Aguilar et al., 2021) However, studies on stakeholder integration,
usefulness of their models by using real Smart Grid data. The integration optimization strategies, and uncertainty for grid stability and efficiency
of computer science, electrical engineering, and data analytics in this highlight the significance of a comprehensive approach to energy
interdisciplinary approach enables a thorough comprehension of the management. Even though using ML and DL is similar, there are dif­
complexity of Smart Grid and aids in the creation of strong forecasting ferences in the techniques used and the level of analysis.
models. While some studies take a more comprehensive approach and
examine the entire system dynamics and integration issues, others focus
on the finer points of certain algorithms. The multifaceted nature of load
8.1. Advantages and limitations forecasting in smart grids is reflected in this diversity. the comparative
analysis emphasizes how diverse load forecasting research in smart grids
The analysis uncovers both advantages and disadvantages in the for ML DL smart energy management is. Although there are certain
methodologies employed for energy management and forecasting. An similarities, including the use of sophisticated algorithms, the area is
important advantage is the utilisation of machine learning (ML) and dynamic and ever-evolving as evidenced by the variations in approaches
deep learning (DL) techniques, which employ advanced algorithms and and focal points. This variety opens the door for future solutions that will
past data to detect patterns and generate precise load forecasts. These be more flexible and successful by fostering a thorough awareness of the
methods are highly proficient in managing extensive datasets and opportunities and problems in smart energy management.
adjusting to dynamic fluctuations in energy usage, leading to more ac­
curate predictions. Ensemble methods improve the accuracy of pre­ 9.1. Synthesizing key concepts
dictions by minimising uncertainty and aggregating the results of
several models. Optimisation approaches enhance the overall efficiency Performance measures are essential for assessing the precision and
of a system by modifying the parameters of a model, hence improving efficacy of machine learning (ML) and deep learning (DL) models since
accuracy and adaptability. they encapsulate important concepts. Quantitative measurements are
Nevertheless, these sophisticated methods are accompanied by used to evaluate forecasting models’ accuracy in predicting future
inherent difficulties. The opaque nature of many ML and DL models events. The MAPE, or refers to absolute percentage error, Root Mean
might obfuscate their decision-making processes, rendering the inter­ Squared Error (RMSE) or RRSE & Mean Squared Error are some exam­
pretation and reliance on forecasts challenging. Data-related challenges ples of these measures. The mean squared error (MSE) is the primary
can provide substantial obstacles; datasets that are biassed, inconsistent, metric used to calculate the average squared disparity between the
or insufficient can compromise the accuracy of load estimates and actual and anticipated data. RMSE, which is derived from MSE, provides
impact the dependability of energy management systems. Moreover, a more understandable statistic by computing the square root and
machine learning (ML) and deep learning (DL) models might require reporting the average magnitude of errors in the projected values.
significant computational resources, which can pose challenges in terms A relative version of RMSE that scales the error in accordance with
of scalability and efficiency, especially in real-time applications. the range of actual values is RRSE, a standardized metric of prediction

3665
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

accuracy. The average deviation between the observed and expected ∑F− 1 ∑F− 1 (l− 1)
values is displayed using a percentage-based metric called the Mean Zi,j = Xi+k,j+lʹ .Wk,lʹ + b(l) (7)
(l) (l)
k=0 lʹ=0
Absolute Percentage Error, or MAPE. Because it expresses forecast ac­
curacy plainly, this statistic is particularly useful when working with Where:
diverse energy consumption scales. Performance metrics, which sum­ Z (i,j)(l) represents the value at location (i,j) of the feature map in
marize key ideas, are crucial for evaluating the accuracy and effective­ layer l. It is calculated by summing the values of the previous layer’s
ness of deep learning (DL) and machine learning (ML) models. feature map, denoted as Z (i+k, j+l)(l-1), within a specific range.
The accuracy with which forecasting models can predict future The value at location (i+k, j+l′) in layer l-1 is denoted as (l-1) and is
events is assessed using a variety of quantitative metrics, such as Mean obtained by multiplying the input at position (i, j′) in layer l-1 with the
Squared Error (MSE), Root Mean Squared Error (RMSE), Relative Root weight Wk,l′.
Mean Squared Error (RRSE), and Mean Absolute Percentage Error (l) Represents the weight at the point (k, l′) of the filter in layer l. b (l)
(MAPE). The mean squared error (MSE) is the primary metric used to represents the bias term in layer l.
calculate the average squared disparity between the actual and antici­ The size of the filter is denoted by F.
pated data. RMSE, which is derived from MSE, provides a more under­ LSTM equation
standable statistic by computing the square root and reporting the
ot = σ (Wio • xt + bio + Who • ht− 1 + bho ) (8)
average magnitude of errors in the projected values. A standardized
indicator of prediction accuracy, RRSE, is a relative form of RMSE that (xt) represents the input at time (t), (h_{t-1}) represents the prior
scales the error according to the actual values’ range. hidden state. (W) and (b) are the weights and biases respectively.
A percentage-based metric known as the Mean Absolute Percentage (sigma) refers to the sigmoid function.
Error, or MAPE, shows the average variation between the observed and Ensemble techniques, which integrate predictions from various al­
expected values. In particular, this statistic is especially helpful when gorithms to resolve uncertainties and improve forecasting reliability, are
dealing with different energy consumption scales because it clearly essential models. Examples of these techniques are Random Forests and
communicates forecast accuracy. A thorough understanding of model Gradient Boosting. Due to their ability to mitigate the constraints of
performance that strikes a balance between precision and interpret­ individual models, these models enhance the robustness of energy
ability is produced by combining these crucial variables. Researchers management systems (Xiong et al., 2022).
often combine these measures to perform a detailed analysis of their Given an input sample X, the projected output Y is obtained by
forecasting models and provide a more comprehensive picture of pre­ combining the predictions Yi from each individual decision tree i in the
diction accuracy. forest. In regression tasks, the average prediction might serve as the
This enables them to consider metrics that are relative as well as benchmark.
absolute. These metrics serve as both benchmarks for researchers and
̂ = 1 n Yi
∑
assistance to practitioners in selecting models that satisfy specific fore­ Y (9)
n i=1
casting objectives. Combining the findings of the MSE, RMSE, RRSE, and
MAPE facilitates a thorough evaluation. This demonstrates the necessity The symbol "^" represents the predicted value.
of developing models that not only minimize errors but also account for The variable n represents the quantity of trees present in the Random
the size and context of the energy forecasting issue. All of these core Forest.
concepts, which are firmly grounded in performance metrics, highlight Yi represents the forecasted value made by the i-th decision tree.
the importance of careful quantitative evaluation in advancing the field Gradient Boosting Equation:
of load forecasting. The emphasis on accuracy metrics draws attention to ̂ = FM (x)
Y (10)
the continuous work being done to enhance and optimize ML and DL
models for more trustworthy and effective smart energy management in The symbol (F_M(x)) denotes the model obtained at iteration (M) in
the rapidly evolving field of smart grids. the process of Gradient Boosting. An ensemble of weak learners is suc­
cessively updated to minimize the loss function (L), which quantifies the
9.2. Frameworks and models discrepancy between predicted and true values. During each iteration
(m), a weak learner (h_m(x)) is trained using the negative gradient of the
The methodological groundwork for creating and putting into loss function. The step size (gamma_m) defines how much (h_m(x))
practice intelligent energy management systems is provided by frame­ contributes to the overall model (F_m(x)). The ultimate forecast (hat{y})
works and models in load forecasting for smart grid for smart energy is the result of the boosted model following (M) iterations.
management. These models rely heavily on machine learning (ML) as The use of optimization frameworks, which make use of meta­
well as deep learning (DL) frameworks, which provide a variety of heuristic algorithms such as particle swarm optimization or genetic al­
methods for handling the intricacies of load forecasting. TensorFlow, gorithms, is essential for optimizing model parameters and achieving
PyTorch, and scikit-learn are popular machine learning frameworks that higher efficiency. These frameworks match the unique needs of Smart
offer a powerful toolkit for performing ensemble techniques, time series Grids with the ML and DL models, optimizing their performance. The
analysis, and regression. These frameworks make it easier to create models and frameworks used in load forecasting for smart grids are a
models that can adjust dynamically to shifts in the patterns of energy complex toolbox that includes optimization approaches, ensemble
consumption, maximizing the efficiency and dependability of grid methods, machine learning, and deep learning. By combining these
operations. methodological techniques, we can create intelligent and flexible energy
DL models are essential parts of load forecasting frameworks, and management systems that will lead to more sustainable, dependable,
they are frequently implemented using neural network architectures.
LSTM networks and CNNs are attractive candidates due to their capa­
bility to extract spatial patterns and temporal dependencies from energy Table 3
Comparison table of existing work.
data. These deep learning techniques enhance the accuracy of load
predictions by discovering intricate connections within the data. The Model Performance References
essence of CNNs lies in the convolutional operation, where filters are Long Short-Term Memory (LSTM) Accuracy 0 f 84 % (Rabie et al., 2024)
applied to input data, generating feature maps. CNN model Accuracy of 94.5 % (Zhang et al., 2023)
CNN model Accuracy of 98.83 % (Mohsin et al., 2023)
LSTM model Accuracy of 77.86 % (Yanmei et al., 2024)

3666
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

and effective energy sources in the future. will be aided by interdisciplinary research that takes into account both
Table 3 titled "Comparison Table of Existing Work" presents a sum­ technical and socioeconomic factors.
mary of the performance of different models used in a specific research Future studies can advance load forecasting in smart grids and make
context, showcasing their accuracy. The first entry is the Long Short- energy systems more resilient, sustainable, and able to adapt to the
Term Memory (LSTM) model, which achieved an accuracy of 84 %. changing demands of society by tackling these topics. These areas of
LSTM models are known for their ability to handle sequential data and research could influence smart energy management in the future and
long-term dependencies, making them suitable for tasks such as time help the world’s energy systems become cleaner and more efficient.
series prediction and natural language processing.
Next, the table lists two entries for Convolutional Neural Network 10.1. Technology and methodology advancements
(CNN) models. The first CNN model achieved an accuracy of 94.5 %.
CNNs are typically used in image and video recognition due to their The field of energy forecasting is changing due to significant tech­
proficiency in capturing spatial hierarchies in data. The second CNN nological and methodological breakthroughs in load forecasting in
model significantly outperformed the first, with an accuracy of 98.83 %. Smart Grid for Smart Energy Management. Advances in machine
This high accuracy indicates the model’s superior performance in the learning (ML) and deep learning (DL) technologies are improving the
given task, potentially due to more advanced architecture or better accuracy and flexibility of load forecasting models. Neural networks and
training data. The final entry is another LSTM model, which had a lower ensemble techniques are examples of advanced machine learning algo­
accuracy of 77.86 %. This variation in performance between LSTM rithms that are constantly developing to handle complex interactions
models may be due to differences in dataset characteristics, model seen in energy consumption data.
tuning, or implementation details. The table effectively highlights the Predictions become more accurate when deep learning architectures,
comparative performance of LSTM and CNN models, with CNN models such as Long Short-Term Memory (LSTM) networks and Convolutional
generally showing higher accuracy in this context. Neural Networks (CNN), are integrated. This allows for the modeling of
intricate temporal and spatial patterns. Beyond algorithmic complexity,
10. Future directions methodological advances include areas such as explainable AI. There is a
current endeavor to improve the interpretability of models by increasing
Future research can be classified under six distinct categories which the transparency of their decision-making procedures.
are explained below. Gaining acceptability and trust is contingent upon this, especially in
situations where stakeholders must comprehend the reasoning behind
1) Intelligible AI for Forecasting Loads energy estimates. Likewise, real-time load forecasting capabilities have
One major obstacle that still needs to be overcome is making been added with the rise of edge computing. Faster decision-making is
machine learning and deep learning models interpretable. Subse­ made possible by processing data closer to the source, which is crucial
quent investigations may concentrate on creating and refining for controlling the dynamic and quickly shifting nature of energy con­
explainable AI methods in order to offer perceptions into the sumption. Forecasting models are further refined through the inclusion
reasoning processes of intricate models. This would boost confidence of optimization techniques through the use of metaheuristic algorithms.
and make it easier for these models to be used in more practical By bringing energy management systems into compliance with the
contexts. unique needs and limitations of smart grids, this optimization increases
2) Renewable Energy Source Integration the effectiveness of those systems. It represents a paradigm-shifting
Studying how to successfully include renewable energy sources period in energy forecasting due to the convergence of technology and
into load forecasting models is important as the world progresses methodological advances. These developments contribute to the devel­
toward a more sustainable energy future. For grid stability and ef­ opment of intelligent and sustainable energy infrastructures by
ficiency, it is essential to address the intermittent character of increasing prediction accuracy and efficiency and opening the door for
renewable energy and optimize its integration with traditional more transparent, flexible, and real-time energy management systems.
sources.
3) Smart Grid Cybersecurity 10.2. Predictions for the futurework
As Smart Grids depend more and more on digital technology,
cybersecurity becomes critical. Subsequent investigations must to In the realm of energy management, a multitude of projections shape
focus on creating strong cybersecurity defenses against possible a future rich with intriguing possibilities. The evolution of machine
cyberattacks on Smart Grids, guaranteeing the confidentiality and learning algorithms holds the promise of yielding forecasting models
integrity of data and energy infrastructure. that are progressively more precise and adaptable, capable of navigating
4) Advanced Computing for Instantaneous Prediction a spectrum of dynamic energy use patterns. Addressing the interpret­
One interesting area of investigation might be the use of edge ability challenge, Explainable AI is poised to gain popularity, fostering
computing in load forecasting. By processing data closer to the wider acceptance of intricate models and resolving the intricacies
source, edge computing lowers latency and facilitates real-time de­ associated with their outputs.
cision-making. This can be especially helpful when it comes to pat­ Real-time decision-making, a pivotal aspect in steering the fluctu­
terns of energy consumption that are dynamic and changing quickly. ating demands of energy systems, is on the horizon with the integration
5) Energy Management Systems Focused on Humans of edge computing. This integration will empower swift responses to
Incorporating user preferences and behavior into energy man­ changing energy needs, ensuring efficiency and resilience in energy
agement systems may result in more individualized and human- management. Emphasizing a comprehensive approach to smart energy
centered strategies. Subsequent investigations could examine the management, interdisciplinary collaboration among data scientists, en­
integration of social sciences and behavioral economics to enhance ergy experts, and policymakers is slated to become more prevalent.
comprehension and modeling of the human elements impacting en­ This collaborative effort will be instrumental in navigating the
ergy usage. complexities of the evolving energy landscape. Anticipated innovations
6) Interaction across Domains in forecasting models are set to revolutionize the field by incorporating
user preferences and behaviors. This shift towards individualized and
Research collaborations between data scientists, energy specialists, human-centered energy solutions aligns with a vision of a more
and policymakers may help close the gap between policy implementa­ personalized and responsive energy management system. Collectively,
tion and technology improvements. The development of Smart Grids these forecasts paint a compelling vision of the future – one where

3667
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

technology-driven, transparent, and responsive energy management References

systems play a pivotal role in establishing resilient and sustainable smart
grids. The integration of cutting-edge technologies and collaborative A. Kermani et al., “Energy Management System for Smart Grid in the Presence of Energy
Storage and Photovoltaic Systems,” vol. 2023, 2023.
approaches is poised to transform the energy sector into a dynamic and Abdolrasol, Maher G.M., et al., 2023. Optimal fuzzy logic controller based PSO for
adaptive ecosystem, capable of meeting the challenges of tomorrow. photovoltaic system. Energy Rep. 9, 427–434.
Aftab, Mohd Asim, et al., 2021. IEC 61850 communication based dual stage load
frequency controller for isolated hybrid microgrid. Int. J. Electr. Power Energy Syst.
11. Conclusions 130, 106909.
Aguilar, J., Garces-Jimenez, A., R-Moreno, M.D., García, R., 2021. A systematic literature
This work provides a distinctive exploration of Deep Learning (DL) review on the use of artificial intelligence in energy self-management in smart
buildings. Renew. Sustain. Energy Rev. vol. 151 (August), 111530. https://doi.org/
and Machine Learning (ML) techniques applied to smart grid load 10.1016/j.rser.2021.111530.
forecasting within energy management systems. Unlike traditional Alazzam, M.B., Alassery, F., Almulihi, A., 2021. A novel smart healthcare monitoring
literature reviews, which primarily classify and analyze individual system using machine learning and the internet of things. Wirel. Commun. Mob.
Comput. vol. 2021. https://doi.org/10.1155/2021/5078799.
studies, this paper adopts a forward-looking approach by not only
Almadhor, A., et al., 2022. Solar power generation in smart cities using an integrated
assessing current methodologies but also proposing novel research di­ machine learning and statistical analysis methods. Int. J. Photo vol. 2022. https://
rections. The review underscores the critical role of AI and ML in doi.org/10.1155/2022/5442304.
advancing load forecasting accuracy and scalability, highlighting their Al-Quraan, M., Khan, A., Centeno, A., Zoha, A., Imran, M.A., Mohjazi, L., 2023.
FedraTrees: a novel computation-communication efficient federated learning
integration into real-world applications through comprehensive evalu­ framework investigated in smart grids. Eng. Appl. Artif. Intell. vol. 124 (September
ations of neural networks, ensemble methods, and probabilistic fore­ 2022), 106654. https://doi.org/10.1016/j.engappai.2023.106654.
casting techniques. Alquthami, T., Zulfiqar, M., Kamran, M., Milyani, A.H., Rasheed, M.B., 2022.
A performance comparison of machine learning algorithms for load forecasting in
The study reveals a dynamic landscape characterized by technical smart grid. IEEE Access vol. 10, 48419–48433. https://doi.org/10.1109/
innovation and interdisciplinary collaboration, emphasizing precision, ACCESS.2022.3171270.
transparency, and sustainability in smart energy management. By sys­ Alsharekh, M.F., Habib, S., Dewi, D.A., Albattah, W., Islam, M., Albahli, S., 2022.
Improving the efficiency of multistep short-term electricity load forecasting via R-
tematically comparing DL and ML techniques against traditional CNN with ML-LSTM. Sensors vol. 22 (18). https://doi.org/10.3390/s22186913.
methods using performance metrics such as MAPE, RMSE, and MSE, the Al-Shetwi, A.Q., Issa, W.K., Aqeil, R.F., Al-Masri, H.M.K., Alzaareer, K., Abdolrasol, M.G.
review provides a robust assessment of their efficacy in addressing M., Abdullah, M.A., 2022. Active power control to mitigate frequency deviations in
large-scale grid-connected PV system using grid-forming single-stage inverters.
challenges like grid stability and demand forecasting. Real-world case Energies 15, 2035.
studies demonstrate how these advanced techniques contribute to Alsirhani, A., Mujib Alshahrani, M., Abukwaik, A., Taloba, A.I., Abd El-Aziz, R.M.,
practical solutions, enhancing the reliability and efficiency of energy Salem, M., 2023. A novel approach to predicting the stability of the smart grid
utilizing MLP-ELM technique. Alex. Eng. J. vol. 74, 495–508. https://doi.org/
systems.
10.1016/j.aej.2023.05.063.
A distinguishing feature of this review is its forward-looking Anantha Krishnan, V., et al., 2022. An IOT innovation of smart solar energy consumption
perspective, which not only examines current methodologies but also analysis and control in micro grid. Int. J. Photo vol. 2022. https://doi.org/10.1155/
proposes future research directions. By identifying key challenges such 2022/7506237.
Asiri, M.M., Aldehim, G., Alotaibi, F.A., Alnfiai, M.M., Assiri, M., Mahmud, A., 2024.
as data quality issues, scalability concerns, and the need for interpret­ Short-term load forecasting in smart grids using hybrid deep learning. IEEE Access
ability, this review sets the stage for addressing these issues in future vol. 12, 23504–23513. https://doi.org/10.1109/ACCESS.2024.3358182.
studies. Moreover, it emphasizes the integration of AI and ML into B. Islam, M. Rasheed, and S.F. Ahmed, “Review of Short-Term Load Forecasting for Smart
Grids Using Deep Neural Networks and Metaheuristic Methods, vol. 2022, 2022.
practical applications within energy management, showcasing their Balasubramaniam, A., Balasubramaniam, T., Paul, A., Seo, H.C., 2022. Electric vehicle
potential to optimize grid stability, enhance demand prediction, and usage pattern analysis using nonnegative matrix factorization in renewable ev-smart
manage uncertainties effectively. charging grid environment. Math. Probl. Eng. vol. 2022. https://doi.org/10.1155/
2022/9365214.
Moving forward, the review suggests exploring hybrid models, Barik, A.K., Das, D.C., Latif, A., 2021. Optimal voltage–frequency regulation in
refining prediction frameworks, and developing standardized evalua­ distributed sustainable energy-based hybrid microgrids with integrated resource
tion metrics to further advance the field of load forecasting in smart planning. Energies 14, 2735.
Bolla, S., Anandan, R., Thanappan, S., 2022. Weather forecasting method from sensor
grids. These insights not only contribute to academic understanding but transmitted data for smart cities using IoT. Sci. Program. vol. 2022. https://doi.org/
also provide a roadmap for implementing robust and efficient energy 10.1155/2022/1426575.
management strategies in real-world scenarios Botchkarev, A., 2019. A new typology design of performance metrics to measure errors in
machine learning regression algorithms. Interdiscip. J. Inf., Knowl., Manag. vol. 14,
45–76. https://doi.org/10.28945/4184.
CRediT authorship contribution statement Boum, A.T., Junior, V., Kakeu, F., Mbey, C.F., 2022. Photovoltaic power generation
forecasting using a novel hybrid intelligent model in smart grid, vol. 2022.
Subhasish Deb: Writing – review & editing, Writing – original draft, Chatterjee, D., Biswas, P.K., Sain, C., Roy, A., Islam, M.M., 2024. SVM classifier based
energy management strategy for dual-source fuel cell hybrid electric vehicles. J. Eng.
Conceptualization. Biswajit Biswal: Writing – review & editing, 2024, e12392.
Writing – original draft, Conceptualization. Taha Selim Ustun: Writing Chauhan, A., Upadhyay, S., Khan, M.T., 2021. Performance investigation of a solar
– review & editing, Writing – original draft, Conceptualization. Subir photovoltaic/diesel generator based hybrid system with cycle charging strategy
using BBO algorithm. Sustainability 13, 8048.
Datta: Writing – review & editing, Writing – original draft, Conceptu­ Chen, Y., et al., 2023. Evaluation of machine learning models for smart grid parameters:
alization. Umit Cali: Writing – review & editing, Writing – original performance analysis of ARIMA and Bi-LSTM. ” Sustain vol. 15 (11). https://doi.org/
draft, Conceptualization. 10.3390/su15118555.
Complementary, E., Medicine, A., 2023. Retracted. A Smart Solut. Cancer Patient Monit.
vol. 2022.
Declaration of Competing Interest Das, S.S., Das, A., Dawn, S., Gope, S., 2022b. A Joint Scheduling Strategy for Wind and
Solar Photovoltaic Systems to Grasp Imbalance Cost in Competitive Market.
Sustainability 14, 5005.
The authors declare that they have no known competing financial Das, A., Dawn, S., Gope, S., 2022a. A risk curtailment strategy for solar PV-battery
interests or personal relationships that could have appeared to influence integrated competitive power system. Electronics 11, 1251.
the work reported in this paper. Das, R.P., Samal, T.K., Luhach, A.K., 2023. An energy efficient evolutionary approach for
smart city-based iot applications. Math. Probl. Eng. vol. 2023. https://doi.org/
10.1155/2023/9937949.
Data availability Dawn, S., Gope, S., Das, S.S., 2021. Social welfare maximization of competitive
congested power market considering wind farm and pumped hydroelectric storage
No data was used for the research described in the article. system. Electronics 10, 2611.
Design and Implementation of Maiden Dual-Level Controller for Ameliorating Frequency
Control in a Hybrid Microgrid. Energies 2021a, 14, 2418.

3668
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

Dey, Bishal, et al., 2023. Forecasting ethanol demand in India to meet future blending Liu, R., Liu, L., 2022. Research on intelligent recognition and management of smart city
targets: a comparison of ARIMA and various regression models. Energy Rep. 9, based on machine vision. J. Sens., Vol. 2022. https://doi.org/10.1155/2022/
411–418. 6605532.
Dey, P.P., Das, D.C., Latif, A., Hussain, S.M.S., 2020. Active power management of virtual Livieris, I.E., Pintelas, E., Pintelas, P., 2020. A CNN–LSTM model for gold price time-
power plant under penetration of central receiver solar thermal-wind using butterfly series forecasting (Dec). Neural Comput. Appl. vol. 32 (23), 17351–17360. https://
optimization technique. Sustainability 12, 6979. doi.org/10.1007/S00521-020-04867-X/METRICS.
Distributed Energy Resources (DER) object modeling with IEC 61850–7–420," AUPEC Ma, Y., Chen, X., Wang, L., Yang, J., 2021. Study on smart home energy management
2011, Brisbane, QLD, Australia, 2011, pp. 1-6. system based on artificial intelligence. J. Sens. Vol. 2021. https://doi.org/10.1155/
Duangsuwan, S., Juengkittikul, P., Myint Maw, M., 2021. Path loss characterization 2021/9101453.
using machine learning models for GS-to-UAV-enabled communication in smart Maghraoui, A.El, Hammouch, F.E., Ledmaoui, Y., Chebak, A., 2022. Smart energy
farming scenarios. Int. J. Antennas Propag. vol. 2021. https://doi.org/10.1155/ management system: A comparative study of energy consumption prediction
2021/5524709. algorithms for a hotel building. Proc. - 2022 IEEE 4th Glob. Power, Energy Commun.
Farooq, Z., Rahman, A., 2022. Power generation control of renewable energy based Conf. GPECOM 2022, pp. 529–534. https://doi.org/10.1109/
hybrid deregulated power system. Energies 15, 517. GPECOM55404.2022.9815807.
Fekri, M.N., Grolinger, K., Mir, S., 2023. Asynchronous adaptive federated learning for Mehmood, M.Y., et al., 2021. Edge Computing for IoT-Enabled Smart Grid. Secur.
distributed load forecasting with smart meter data. Int. J. Electr. Power Energy Syst. Commun. Netw. vol. 2021. https://doi.org/10.1155/2021/5524025.
vol. 153 (May), 109285. https://doi.org/10.1016/j.ijepes.2023.109285. Mohsin, S.M., Khan, F.A., Mustafa, K., 2023. Short. -Term. Electr. -Load. Forecast. Smart
G. M. Abdolrasol, et. al. Energy Management Scheduling for Microgrids in the Virtual Grids 1–16.
Power Plant System Using Artificial Neural Networks. Energies 2021b, 14, 6507. Muzumdar, A., Modi, C., Vyjayanthi, C., 2020. An efficient regional short-term load
Ganesan, A., Paul, A., Seo, H.C., 2022. Elderly people activity recognition in smart grid forecasting model for smart grid energy management. IECON Proc. (Ind. Electron.
monitoring environment. Math. Probl. Eng. vol. 2022 (Ml). https://doi.org/ Conf. vol. 2020-Octob, 2089–2094. https://doi.org/10.1109/
10.1155/2022/9540033. IECON43393.2020.9254736.
F. Ghislain et al., “A Novel Smart Method for State Estimation in a Smart Grid Using Nadeem, F., Aftab, M.A., Hussain, S.M.S., Ali, I., Tiwari, P.K., Goswami, A.K., 2019.
Smart Meter Data,” vol. 2022, 2022. Virtual power plant management in smart grids with XMPP based IEC 61850
H. You, “Safe Operation Management of Urban Smart Grid Based on Deep Learning,” vol. communication. Energies 12, 2398.
2022, 2022. Nagarajan, G., et al., 2022. A novel edge-based trust management system for the smart
Hamoudi, Y., Amimeur, H., Aouzellag, D., 2023. Hyperparameter bayesian optimization city environment using eigenvector analysis. J. Healthc. Eng. vol. 2022. https://doi.
of gaussian process regression applied in speed-sensorless predictive torque control org/10.1155/2022/5625897.
of an autonomous wind energy conversion system. Energies 16, 4738. Nayagam, V.S., et al., 2022. Deep learning model on energy management in grid-
Haq, E.U., Pei, C., Zhang, R., Jianjun, H., Ahmad, F., 2023. Electricity-theft detection for connected solar systems. Int. J. Photo vol. 2022. https://doi.org/10.1155/2022/
smart grid security using smart meter data: a deep-CNN based approach. Energy Rep. 6371182.
vol. 9, 634–643. https://doi.org/10.1016/j.egyr.2022.11.072. Onteru, R.R., Vuddanti, S., 2023. A novel framework for machine intelligence aided
Hasan, M.K., et al., 2022. Blockchain Technology on Smart Grid, Energy Trading, and Big smarter energy management in residential scenario. Meas. Sens. vol. 28 (May),
Data: security Issues, Challenges, and Recommendations. Wirel. Commun. Mob. 100855. https://doi.org/10.1016/j.measen.2023.100855.
Comput. vol. 2022. https://doi.org/10.1155/2022/9065768. Patil, G.S., Mulla, A., Dawn, S., 2022. Profit maximization with imbalance cost
Healthcare Engineering, J.O., 2023. Retracted: smart medical prediction for guidance: a improvement by solar PV-battery hybrid system in deregulated power market.
mechanism study of machine learning. J. Healthc. Eng. vol. 2023, 9871325. https:// Energies 15, 5290.
doi.org/10.1155/2023/9871325. Pattanaik, S.S., Sahoo, A.K., Panda, R., Dawn, S., 2024. Optimal power allocation of
Hussain, Israfil, et al., 2022. Active power control of autonomous hybrid power system battery energy storage system (BESS) using dense LSTM in active distribution
using two degree of freedom PID controller. Energy Rep. 8, 973–981. network. Energy Storage 6 (1), e529.
Hussain, S.S., Aftab, M.A., Ali, I., 2020a. IEC 61850 based energy management system Prasath, S., et al., 2022. Development of microbiology plantation-based multimodal
using plug-in electric vehicles and distributed generators during emergencies. Electr. segmentation for smart garden using machine learning. Adv. Mater. Sci. Eng. vol.
Power Energy Syst. 119, 105873. 2022. https://doi.org/10.1155/2022/1066535.
Hussain, S.M.S., Aftab, M.A., F. Nadeem, I., 2020a. Optimal energy routing in microgrids Rabie, A.H., I. Saleh, A., Elkhalik, S.H.A., Takieldeen, A.E., 2024. An optimum load
with IEC 61850 based energy routers (June). IEEE Trans. Ind. Electron. vol. 67 (6), forecasting Strategy (OLFS) for smart grids based on artificial intelligence.
5161–5169. Technologies vol. 12 (2). https://doi.org/10.3390/technologies12020019.
Hussain, I., Das, D.C., Sinha, N., Latif, A., Hussain, S.M.S., 2020b. Performance Ranjan, Sudhanshu, et al., 2021b. Simultaneous analysis of frequency and voltage
assessment of an islanded hybrid power system with different storage combinations control of the interconnected hybrid power system in presence of FACTS devices and
using an FPA-tuned two-degree-of-freedom (2DOF) controller. Energies 13, 5610. demand response scheme. Energy Rep. 7, 7445–7459.
Ibrar, M., et al., 2022. A machine learning-based model for stability prediction of Ranjan, Sudhanshu, et al., 2021a. Voltage stability assessment of isolated hybrid dish-
decentralized power grid linked with renewable energy resources. Wirel. Commun. stirling solar thermal-diesel microgrid with STATCOM using mine blast algorithm.
Mob. Comput. vol. 2022. https://doi.org/10.1155/2022/2697303. Electr. Power Syst. Res. 196, 107239.
Jiang, W., et al., 2023. A stochastic online forecast-and-optimize framework for real-time Refaai, M.R.A., Vonteddu, S.N.R., Nunna, P.K., Kumar, P.S., Anbu, C., Markos, M., 2022.
energy dispatch in virtual power plants under uncertainty. Int. Conf. Inf. Knowl. Energy management prediction in hybrid pv-battery systems using deep learning
Manag. Proc. (2), 4646–4652. https://doi.org/10.1145/3583780.3614653. architecture. Int. J. Photo vol. 2022. https://doi.org/10.1155/2022/6844853.
Kamrul, M., Aliyu, R., Islam, S., Reddy, T., Safie, N., 2024. Review article A review on Safiullah, Sheikh, et al., 2022. Robust frequency–voltage stabilization scheme for multi-
machine learning techniques for secured cyber-physical systems in smart grid area power systems incorporated with EVs and renewable generations using AI based
networks. Energy Rep. vol. 11 (July 2023), 1268–1290. https://doi.org/10.1016/j. modified disturbance rejection controller. Energy Rep. 8, 12186–12202.
egyr.2023.12.040. Sahoo, A., Hota, P.K., Sahu, P.R., Alsaif, F., Alsulamy, S., 2023. Optimal congestion
Khan, M.F., et al., 2021b. A review of big data resource management: using smart grid management with FACTS devices for optimal power dispatch in the deregulated
systems as a case study. Wirel. Commun. Mob. Comput. vol. 2021. https://doi.org/ electricity market. Axioms 12, 614.
10.1155/2021/3740476. Singh, P., Khan, B., 2017. Smart microgrid energy management using a novel artificial
Khan, M.F., et al., 2021a. An iomt-enabled smart healthcare model to monitor elderly shark optimization. Complexity vol. 2017. https://doi.org/10.1155/2017/2158926.
people using machine learning technique. Comput. Intell. Neurosci. vol. 2021. Singh, N.K., Koley, C., Gope, S., Dawn, S., 2021. An economic risk analysis in wind and
https://doi.org/10.1155/2021/2487759. pumped hydro energy storage integrated power system using meta-heuristic
Khan, N., et al., 2022. Energy management systems using smart grids: an exhaustive algorithm. Sustainability 13, 13542.
parametric comprehensive analysis of existing trends, significance, opportunities, Sodagudi, S., et al., 2022. Renewable energy based smart grid construction using hybrid
and challenges. Int. Trans. Electr. Energy Syst. vol. 2022. https://doi.org/10.1155/ design in control system with enhancing of energy efficiency of electronic converters
2022/3358795. for power electronic in electric vehicles. Int. Trans. Electr. Energy Syst. vol. 2022.
Kowalek, P., 2019. Classification of diffusion modes in single-particle tracking data: https://doi.org/10.1155/2022/2986605.
feature-based versus deep-learning approach. Published 20 September Phys. Rev. E Srisomboon, K., Dindam, T., Lee, W., 2021. Empowered hybrid parent selection for
100, 032410. improving network lifetime, PDR, and latency in smart grid. Math. Probl. Eng. vol.
Kumar, N., Singh, G., Kebede, H., 2023. An optimized framework of the integrated 2021. https://doi.org/10.1155/2021/5551152.
renewable energy and power quality model for the smart grid. Int. Trans. Electr. Srivastava, Ashtabhuj Kumar, et al., 2022. Analysis of GOA optimized two-stage
Energy Syst. vol. 2023 (i). https://doi.org/10.1155/2023/6769690. controller for frequency regulation of grid integrated virtual power plant. Energy
Latif, A., Hussain, S.M.S., Das, D.C., 2021. Optimization of Two-Stage IPD-(1+I) Rep. 8, 493–500.
Controllers for Frequency Regulation of Sustainable Energy Based Hybrid Microgrid Su, Q., 2022. Smart teaching design mode based on machine learning and its effect
Network. Electronics 10, 919. evaluation. Math. Probl. Eng. vol. 2022, 9837240. https://doi.org/10.1155/2022/
Latif, A., Paul, M., Das, D.C., 2020. Price based demand response for optimal frequency 9019339.
stabilization in ORC solar thermal based isolated hybrid microgrid under salp swarm Sujatha, M., et al., 2022. IoT and machine learning-based smart automation system for
technique. Electronics 9, 2209. industry 4.0 using robotics and sensors. J. Nanomater., Vol. 2022. https://doi.org/
Li, X., Zhou, C., Liang, Z., Yu, Q., Chen, X., He, Z., 2022. UCB-based route and power 10.1155/2022/6807585.
selection optimization for SDN-enabled industrial IoT in smart grid. Wirel. Commun. T. Yang, T. Huang, H. Zhang, P. Li, and C. Xiong, “Demand Response Management
Mob. Comput. vol. 2022. https://doi.org/10.1155/2022/7424854. Research Based on Cognitive Radio for Smart Grid,” vol. 2020, 2020.

3669
B. Biswal et al. Energy Reports 12 (2024) 3654–3670

Tang, F., Pang, J., Cheng, K., Gong, Q., 2021. Multiauthority traceable ring signature Xu, Z., Gao, Y., Hussain, M., Cheng, P., 2020. Demand side management for smart grid
scheme for smart grid based on blockchain. Wirel. Commun. Mob. Comput. vol. based on smart home appliances with renewable energy sources and an energy
2021. https://doi.org/10.1155/2021/5566430. storage system. Math. Probl. Eng. vol. 2020, 14–22. https://doi.org/10.1155/2020/
Teekaraman, Y., Kirpichnikova, I., Manoharan, H., Kuppusamy, R., Radhakrishnan, A., 9545439.
2022a. Analysis of smart grid using multimedia sensor networks with effective Xu, Y., Hu, G., You, L., Cao, C., 2021. A novel machine learning-based analysis model for
resource allocation. Math. Probl. Eng. vol. 2022. https://doi.org/10.1155/2022/ smart contract vulnerability. Secur. Commun. Netw. vol. 2021 (June 2016). https://
1854365. doi.org/10.1155/2021/5798033.
Teekaraman, Y., Kirpichnikova, I., Manoharan, H., Kuppusamy, R., Angadi, R.V., Yadav, A.K., Malik, H., 2021. Case Study of Grid-Connected Photovoltaic Power System
Thelkar, A.R., 2022b. Diminution of smart grid with renewable sources using Installed at Monthly Optimum Tilt Angles for Different Climatic Zones in India. IEEE
support vector machines for identification of regression losses in large-scale systems. Access vol. 9, 60077–60088.
Wirel. Commun. Mob. Comput. vol. 2022. https://doi.org/10.1155/2022/6942029. Yang, X., Xu, Y., Ma, C., Yao, T., Xu, L., 2023. Integrated energy management of smart
Torres, P.J.R., García, C.G., Puig, M.F.S., Franco, M., Izquierdo, S.K., 2023. grids in smart cities based on PSO scheduling models. Adv. Civ. Eng. vol. 2023.
Reinforcement learning with probabilistic boolean networks in smart grid models. https://doi.org/10.1155/2023/5794002.
Stud. Syst. Decis. Control vol. 464, 215–224. https://doi.org/10.1007/978-3-031- Yanmei, J., et al., 2024. Enhanced neighborhood node graph neural networks for load
26361-3_19. forecasting in smart grid. Int. J. Mach. Learn. Cybern. vol. 15 (1), 129–148. https://
Ulutas, A., Altas, I.H., Onen, A., 2020. Neuro-fuzzy-based model predictive energy doi.org/10.1007/s13042-023-01796-8.
management for grid connected microgrids. Electronics 9, 900. Yao, Z., Zhang, T., Wang, Q., Zhao, Y., 2022. Short-term power load forecasting of
Ustun, T.S., Hussain, S.M.S., Orihara, D., Iioka, D., 2022. IEC 61850 modeling of an AGC integrated energy system based on attention-CNN-DBILSTM. Math. Probl. Eng. vol.
dispatching scheme for mitigation of short-term power flow variations. Energy Rep. 2022. https://doi.org/10.1155/2022/1075698.
vol. 8 (1), 381–391. Yarar, N., Yagci, M., Bahceci, S., 2023. Artificial neural networks based harmonics
Ustun, T.S., Hussain, S.M.S., 2020. IEC 61850 Modeling of UPFC and XMPP estimation for real university microgrids using hourly solar irradiation and
Communication for Power Management in Microgrids. IEEE Access vol. 8, temperature data (March). Energy Nexus Volume 9, 100172 (March).
141696–141704. You, X., Yan, G., Yang, Z., 2022. New method for high water pressure tunnel design
Ustun, T.S., Hussain, S.M.S., Syed, M.H., Dambrauskas, P., 2021a. IEC-61850-Based based on machine learning for smart grid energy management and groundwater
communication for integrated ev management in power systems with renewable environmental balance. Int. Trans. Electr. Energy Syst. vol. 2022. https://doi.org/
penetration. Energies 14, 2493. 10.1155/2022/4912036.
Ustun, T.S., Hussain, S.M.S., Ulutas, A., Onen, A., Roomi, M.M., Mashima, D., 2021a. Yu, D., Ma, Z., Wang, R., 2022. Efficient smart grid load balancing via fog and cloud
Machine learning-based intrusion detection for achieving cybersecurity in smart computing. Math. Probl. Eng. vol. 2022. https://doi.org/10.1155/2022/3151249.
grids using IEC 61850 GOOSE messages. Symmetry 13, 826. Zafar, M.H., et al., 2023. Step towards secure and reliable smart grids in Industry 5.0: a
Ustun, T.S., Hussain, S.M.S., Yavuz, L., Onen, A., 2021b. Artificial intelligence based federated learning assisted hybrid deep learning model for electricity theft detection
intrusion detection system for IEC 61850 sampled values under symmetric and using smart meters. Energy Rep. vol. 10 (August), 3001–3019. https://doi.org/
asymmetric faults. IEEE Access vol. 9, 56486–56495. 10.1016/j.egyr.2023.09.100.
Ustun, T.S., Ozansoy, C., Zayegh, A., 2012. Simulation of communication infrastructure Zhang, D., Jin, X., Shi, P., Chew, X.Y., 2023. Real-time load forecasting model for the
of a centralized microgrid protection system based on IEC 61850-7-420. Proc. IEEE smart grid using bayesian optimized CNN-BiLSTM. Front. Energy Res. vol. 11 (May),
Third Int. Conf. Smart Grid Commun. (SmartGridComm), Tainan, Taiwan, 5–8 1–12. https://doi.org/10.3389/fenrg.2023.1193662.
Novemb. 492–497. Zhang, X., Lu, Z., Lu, M., 2020. Vehicle speed optimized fuzzy energy management for
W. Guo, K. Zhang, X. Wei, and M. Liu, “Short-Term Load Forecasting Method Based on hybrid energy storage system in electric vehicles. Complexity vol. 2020. https://doi.
Deep Reinforcement Learning for Smart Grid,” vol. 2021, 2021. org/10.1155/2020/2073901.
Wang, K., Liang, Z., Xing, N., Li, B., 2023a. Online fault prediction based on collaborative Zhang, W., Yang, W., Chen, C., Li, N., Bao, Z., Luo, M., 2022. Toward privacy-preserving
filtering in smart grid. Math. Probl. Eng. vol. 2023. https://doi.org/10.1155/2023/ blockchain-based electricity auction for V2G networks in the smart grid. Secur.
5555210. Commun. Netw. vol. 2022. https://doi.org/10.1155/2022/6911463.
Wang, L., Zhu, Y., Du, W., Fu, B., Wang, C., Wang, X., 2023b. A novel model-based Zhong, Y., et al., 2021. Distributed blockchain-based authentication and authorization
reinforcement learning for online anomaly detection in smart power grid. Int. Trans. protocol for smart grid. Wirel. Commun. Mob. Comput. vol. 2021 (3). https://doi.
Electr. Energy Syst. vol. 2023. https://doi.org/10.1155/2023/6166738. org/10.1155/2021/5560621.
Wawale, S.G., et al., 2022. Minimizing the error gap in smart framing by forecasting Zhou, T., Shen, J., Ji, S., Ren, Y., Yan, L., 2020. Secure and intelligent energy data
production and demand using ARIMA model. J. Food Qual. vol. 2022. https://doi. management scheme for smart IoT devices. Wirel. Commun. Mob. Comput. vol.
org/10.1155/2022/1139440. 2020. https://doi.org/10.1155/2020/8842885.
Xin, Q., Alazab, M., Díaz, V.G., Montenegro-Marin, C.E., Crespo, R.G., 2022. A deep Zhou, J., Zhao, J., Wang, L., 2022. An energy management strategy of power-split hybrid
learning architecture for power management in smart cities. Energy Rep. vol. 8, electric vehicles using reinforcement learning. Mob. Inf. Syst. vol. 2022. https://doi.
1568–1577. https://doi.org/10.1016/j.egyr.2021.12.053. org/10.1155/2022/9731828.
Xiong, H., et al., 2022. An efficient searchable symmetric encryption scheme for smart Zhu, Q., Lin, H., Wan, C., Xie, Y., Peng, S., 2022. Integrity protection for data aggregation
grid data. Secur. Commun. Netw. vol. 2022. https://doi.org/10.1155/2022/ in smart grid. Secur. Commun. Netw. vol. 2022. https://doi.org/10.1155/2022/
9993963. 2734487.

3670