ELECTRICITY POWER CONSUMPTION

PREDICTION: A REVIEW PAPER

Majda El Barkani
Asmae Nadifi Yousra Chtouki
School of Science and Engineering
School of Science and Engineering School of Science and Engineering
Al Akhawayn University
Al Akhawayn University Al Akhawayn University
Ifrane, Morocco
Ifrane, Morocco Ifrane, Morocco
[Link]@[Link]
[Link]@[Link] [Link]@[Link]
 Abstract—The growing global energy consumption, driven The rise of machine learning (ML) techniques has
by factors such as population growth, industrial expansion, significantly improved electricity consumption forecasting.
and the increasing electrification of sectors like transportation, Models like Artificial Neural Networks (ANN), Support
makes it crucial to have robust models that predict electricity Vector Machines (SVM), and Random Forests (RF) are
usage accurately. This rise in energy demand has made the capable of analyzing large and complex datasets to capture
power systems unable to balance energy supply and demand. non-linear relationships between factors like weather,
Hence, we need accurate electricity forecasting to ensure that economic activity, and historical usage [10][3]. These
the right amount of energy is generated and distributed to
models are particularly useful for short-term load
prevent power outages, minimize energy waste, and optimize
resource use. This is particularly important in modern energy
forecasting, where energy use can fluctuate on a daily or
grids, where the integration of renewable energy sources like even hourly basis [24][9]. In addition, hybrid models, which
solar and wind introduces variability and unpredictability. combine traditional statistical approaches with ML
Nonetheless, electricity load forecasting has been traditionally techniques, have further enhanced accuracy by leveraging
reliant on statistical models like autoregressive integrated the strengths of both methods [9].
moving average (ARIMA) and seasonal autoregressive Despite the advancements, there are still challenges in
integrated moving average (SARIMA) which recently have
electricity consumption forecasting. One of the main issues is
started becoming obsolete with the demand structure of energy
which has changed drastically in the recent past. The
the need for large and high-quality datasets to train machine
emergence of machine learning (ML) and deep learning (DL) learning models. In many regions, access to such data is
methods saved the situation nearly completely. Models like limited, which can reduce the accuracy of the predictions
Artificial Neural Networks (ANNs), Support Vector Machines [24][3]. Moreover, the increasing need to incorporate real-
(SVMs), and Long Short-Term Memory (LSTM) networks time data, particularly as renewable energy sources like
among others are able to understand and work with such
solar and wind become more common, adds another layer of
complex and non-linear data. However, challenges persist,
complexity. These energy sources are variable, so models
including the need for real-time data, enhancing data quality,
and addressing the "black box" nature of some models, which must adapt to changing conditions in real-time. [5][3][4]
can limit interpretability. This paper provides a comprehensive Another challenge is the lack of standardized evaluation
review of the latest methodologies for electricity consumption metrics across different studies, which makes it difficult to
prediction, covering traditional statistical, machine learning, compare the performance of various models. Metrics like
and deep learning approaches. It also examines recent trends, Root Mean Square Error (RMSE) and Mean Absolute
such as the integration of Internet of Things (IoT) devices for
Percentage Error (MAPE) are commonly used, but their
real-time data collection and the use of hybrid forecasting
effectiveness varies depending on the specific context and
models. The review highlights key areas for future research,
aiming to address the ongoing challenges and improve the type of forecast [24]. Additionally, many machine learning
accuracy and reliability of electricity consumption predictions. models, especially deep neural networks, operate as "black
boxes," meaning that their internal decision-making
Keywords— Global energy consumption, Energy demand and processes are difficult to interpret. This lack of transparency
supply, Electricity forecasting, Statistical models, ARIMA, can be a barrier to their use in critical applications, where
SARIMA, Machine learning, Deep learning, ANNs, SVMs, understanding how predictions are made is important [9][5].
LSTM, Real-time data, Data quality, Model interpretability, IoT,
Hybrid forecasting models. To overcome these challenges, future research needs to
focus on improving data availability, developing more
interpretable models, and integrating real-time data into
I. INTRODUCTION forecasting systems [2]. Accurate electricity consumption
Predicting electricity consumption is a critical task in prediction is essential for ensuring reliable and efficient
managing modern energy systems. As global energy demand energy systems as global demand continues to grow [1].
continues to grow due to factors like population increase,
Our survey examines electricity consumption prediction
industrial expansion, and the shift toward electrification in
from several perspectives: techniques, datasets, and real-
sectors such as transportation, the need for accurate
world applications. We review the latest methods for
electricity forecasting has become more pressing. From 2023
predicting electricity consumption while focusing on
to 2025, global electricity demand is expected to increase by
traditional statistical models, machine learning, and deep
around 3% annually, largely driven by emerging economies
learning approaches. Additionally, we discuss the main
such as China, India, and Southeast Asia [1][2]. Accurate
datasets used in this field by looking at their level of detail,
predictions help ensure the balance between supply and
regional focus, and whether they include external factors like
demand, preventing outages, optimizing energy production,
weather or economic data.
and maintaining grid reliability [1][4].
We also explore how these prediction models are applied
In the past, traditional statistical methods like ARIMA
in areas such as smart grid management, demand-side
and exponential smoothing were commonly used to forecast
management, and renewable energy integration. This survey
electricity consumption. These models worked well when
provides a comprehensive overview by combining an
energy demand patterns were relatively stable and linear, but
evaluation of techniques, an analysis of available datasets,
they struggled to account for the increasing complexity of
and a review of practical applications.
modern energy consumption. Factors such as changing
weather patterns, economic fluctuations, and the growing use In addition to our literature review, we contribute to this
of renewable energy have introduced more variability and field by developing a predictive model using the Tetouan’s
unpredictability into electricity consumption [22][9]. Dataset [16]. This dataset includes energy consumption data
collected at 10-minute intervals along with relevant external
factors such as temperature and solar radiation. Our goal is to
build a model that can accurately forecast electricity across different sectors, and what are their specific
consumption in this region. This will provide insights into applications in both traditional and emerging energy
how advanced machine learning techniques can be applied to systems?
real-world energy systems.
6. In the context of this study, how can we develop a
This paper follows a systematic review approach to predictive model based on advanced machine learning
gather and analyze the most recent research on electricity techniques, using real-world data, such as the Tetouan’s
consumption prediction. The review focuses on a variety of energy consumption dataset, to improve electricity
methods, including statistical models like ARIMA and consumption forecasting accuracy?
SARIMA, machine learning techniques like Random Forest
These research questions help focus the review on
and Support Vector Machines (SVM), and deep learning
identifying strengths, limitations, and opportunities for
approaches such as Long Short-Term Memory (LSTM)
improvement in the field of electricity consumption
networks. The sources for this review were gathered from
prediction.
well-known databases, including IEEE Xplore,
ScienceDirect, SpringerLink, and Google Scholar as shown
in Figure 1. The review spans literature published between II. LITERATURE REVIEW
2000 and 2024, concentrating on key developments and The forecasting of electricity usage has been explored
advancements in electricity consumption forecasting. for many decades due to its critical role in balancing supply
To ensure a comprehensive review various relevant and demand, optimizing grid operations, and integrating
papers, reports, and articles were consulted. These sources renewable energy sources. This research dates back to the
cover a wide range of methods used in electricity early 1980s when statistical techniques were used to predict
consumption forecasting. The following key research electricity demand. As power systems become more
questions guided this review: complex and technological advancements grow, and with
the increased integration of renewable energy, advanced
1. What are the recent advancements in electricity
models are required. Thus, various predictive models have
consumption prediction methods, and how do these emerged to overcome challenges related to accuracy, non-
advancements improve the accuracy of forecasts? linearity, and scalability.
2. How do machine learning and deep learning methods A. Early Approaches: Statistical Methods (1980s-2000s)
compare to traditional statistical models in terms of In the 1980s and 1990s, ARIMA model was commonly
used to predict electricity usage. It could capture simple,
predictive ability, scalability, and ease of use? steady trends over time but struggled to handle complex
3. What are the main challenges faced when applying realworld factors, such as seasonal changes, economic shifts,
and weather patterns, which significantly affect electricity
these models in real-world energy systems, especially consumption.
in areas like smart grids, demand-side management, To address the limitations of ARIMA, SARIMA model
and renewable energy integration? was developed in the early 2000s. SARIMA generalized
ARIMA by adding seasonal parameters, useful for electricity
4. What are the future research directions that can help demand forecasting in regions heavily influenced by weather
overcome current challenges, such as issues with data conditions. For instance, the demand in extreme winters or
summers can rise dramatically due to cooling or heating
quality, model interpretability, and scalability? requirements. In a 2018 study, SARIMA was found effective
5. How are these predictive models currently being used
 Figure 1: Articles related to electricity consumption per journal in the last 10 years

in Prorecasting the seasonal patterns of electricity C. Deep Learning and Its Revolutionary Impact (2010s-
consumption in the urban grid of Tetouan [10]. present)
Despite being adequate for short-term forecasts, The 2010s saw significant developments in deep learning
SARIMA struggled with long-range forecasting, especially (DL) models, which advanced electricity consumption
with structural complexities like economic growth, forecasting by capturing long-term dependencies in data.
unforeseen weather patterns, and the growing use of Major deep learning models include:
renewable energy. SARIMA’s linearity limited its • Recurrent Neural Networks (RNNs): RNNs, with
performance in predicting non-linear relations, particularly in their ability to retain data from previous time steps, are
grids with renewable energy integration [2][3]. highly suitable for temporal data. This made RNNs a popular
choice for predicting electricity demand on an hourly or daily
B. The Rise of Machine Learning Models (2000s-2010s) basis.
In the mid-2000s, ML models emerged as alternatives to
• Long Short-Term Memory (LSTM): LSTM networks,
SARIMA and other statistical models. These models were
an improvement on RNNs, are designed to handle long-term
capable of managing non-linear relationships and analyzing
sequences of data, preventing issues like vanishing gradients.
complex interplays between variables such as climate,
A 2019 study found that bidirectional LSTM (BLSTM)
population growth, and economic activities. Key ML models
networks provided more accurate electricity demand
include:
predictions by exploiting both historical and future data [8].
• Random Forests (RF): RF models, particularly in
• Hybrid DL Models: Hybrid deep learning models,
ensemble learning, became popular for medium- and long-
which combine architectures such as inception networks and
term forecasting. In a 2015 study, RF models enhanced
residual networks, became common in the late 2010s. In
forecasting accuracy for household electricity consumption
2020, hybrid models were shown to outperform single
in Tetouan, especially for long-term predictions (6].
models by recognizing both short-term and long-range
• Support Vector Machines (SVM): SVMs have been dependencies, particularly in grids with renewable energy
widely used for short-term electricity demand forecasting [9].
due to their ability to classify complex consumption patterns,
particularly during adverse climatic conditions. In a 2017 D. Specialized Models for Specific Applications (2010s-
study, incorporating weather and price data into SVM present)
models improved short-term prediction performance [12]. Alongside general models, specialized electricity
• Artificial Neural Networks (ANN): ANNs became consumption forecasting models have been developed for
increasingly used as power systems grew in complexity, specific applications. For instance, short-term load
particularly with the advent of smart grids. ANNs could learn forecasting, which requires predicting cyclic trends like daily
non-linear relationships between variables, such as time of or weekly consumption, benefits from periodically correlated
day, temperature, and demand on the grid. In 2017, ANNs models (10].
were applied to predict daily and weekly electricity • Periodically Correlated Models for Industries: In
consumption more effectively than traditional statistical certain industries, energy consumption follows periodic
methods [7]. patterns due to machine operations. In 2018, researchers
addressed short-term load forecasting in industries that
showed peaks and troughs in energy usage. These models
allowed for improved fuel conservation and reduced energy
waste during low-demand periods [12].
• Manufacturing-Specific Models: Energy demand in
manufacturing can fluctuate sharply with production and
non-production cycles. A 2018 study showed that machine
learning models tailored to specific industrial settings helped
manufacturers adjust energy consumption and reduce costs
[12].

E. Comparative Studies and Hybrid Models (2015-

present)
As predictive models evolved, comparative studies
emerged to evaluate their effectiveness. In 2018, researchers
compared RF, Gradient Boosting Machines (GBM), and
SVM against SARIMA. The findings confirmed that
ensemble methods like RF and GBM outperformed linear
models due to their ability to capture non-linear relationships
in high-dimensional data [14].
• Ensemble Learning Techniques: By 2010, ensemble
learning strategies like bagging (used in RF) and boosting
(used in GBM) were already mainstream. These techniques
reduced overfitting and improved accuracy by combining
predictions from multiple models [15].
• Hybrid SARIMA-ANN Models: In the 2020s, hybrid
models combining SARIMA for linear seasonality with
ANN for non-linear behavior gained prominence. These
models were particularly effective for grids with high
renewable energy penetration, where fluctuations in supply
and demand are more pronounced [16].

F. Recent Trends and Future Directions (2020s-present)

 The introduction of Internet of Things (IoT) devices has Another emerging trend is the integration of demand
modernized electricity consumption prediction by enabling response programs, where consumers are incentivized to
real-time data collection at a granular level. For example, adjust their consumption during peak hours. Predictive
smart meters provide detailed consumption data for each models are increasingly being used to forecast how
appliance, significantly improving short-term forecast consumers will respond to such programs, enhancing grid
Figure 2: Timeline for Electricity Consumption Forecasting Models.

accuracy [10]. efficiency [16].

In 2022, research focused on real-time predictive models
that allow utilities to adjust power generation in response to III. APPLICATION AND CASE STUDIES IN ELECTRICITY
fluctuating demand, reducing waste during times of excess or CONSUMPTION PREDICTION
short supply [16]. The integration of big data analytics with
machine learning systems allows long term forecasting with A. Tetouan’s Dataset in Context
greater accuracy, enabling utilities to optimize energy usage The Tetouan dataset offers a source of information for
and anticipate demand spikes [10]. the study and forecasting of electricity use within an urban
environment owing to its richness and high temporal
In addition, as grids become more decentralized with
resolution. This dataset was collected during the year 2017
solar and wind energy, predictive models are evolving to
by the SCADA system run by Amendis, a public service
become more flexible. Adaptive machine learning
provider in Tetouan. In particular the data consist of 10-
algorithms, which adjust based on real-time IoT data, have
minute periods of electricity consumption records for three
proven more effective than static models in grids with high
main Tariff Areas in the city; Quads, Smir and Boussafou
renewable energy penetration [17].
which provide high granularity in observing short-term
Blockchain technology is also transforming energy changes in energy usage. Each of these zones is
markets. Surplus electricity produced from renewable characterized by different demand profiles, which helps to
sources can now be traded directly between consumers,
adding complexity to electricity usage predictions. Predictive
models are being adapted to incorporate both central utility understand the geographic distribution of power
generation and peer-to-peer energy trading [18]. requirements within the city of Tetouan.

This dataset does not only include power consumption Barring these factors, the Tetouan dataset is also well-
details, but other climatic factors such as temperature, in detailed hence it helps in the evaluation of many models
degree Celsius, humidity expressed as a percentage and wind including the Random Forest (RF) and eXtreme Gradient
speed in kilometers per hour which are significant Boosting (XGBoost) models for these models are well suited
determinants of energy forecasting since energy consumption to the nonlinear dynamics present in energy consumption
patterns fluctuates with these variables. For instance, an forecasting as well as optimistic forecasting results. There is
increase in temperature, especially in places with very hot proof of this dataset being usable in real life; this dataset
summers, is usually associated with increased electricity could support reallocation of city-wide energy distribution
usage due to air conditioning systems [17]. Other specific using the machine learning models to forecast demand which
characteristics that are included in this dataset ‘General will then enhance the efficiency and reliability of the power
Diffuse Flows’ and ‘Diffuse Flows’ represent the air or water grid system in Tetouan. Such information is particularly
movements in the city, which also influence energy usage in useful in the area of urban energy management where such
an indirect context for instance by creating a need for cooling demand forecasting helps in appropriate investment decision
or ventilation. making towards meeting that demand while also considering
efficient use of resources.
B. Case Studies of Similar Datasets purposes. In addition, these studies also confirm that
All over the world, cities have conducted research on the complex and non-linear patterns in electricity data can be
use of similar datasets, and Tétouan is no exception. For processed by the machine learning models basing the
example, The Building Energy Dataset for New York city analysis on Random Forest, SVM or LSTM Latent Models.
consists of energy consumption figures and the ambient
conditions like temperature and dust. The researchers in this C. Comparative Insights
case studied this data and used Random Forest and Support Table 1 below summarizes key insights and comparisons
Vector Machines (SVM) machine learning algorithms to of studies using datasets like Tetouan’s. It highlights the
model the consumption trends. The results indicated that the purpose of each study, the dataset’s characteristics, the
forecasting models were enhanced by the use of weather models applied, the contributions made, and any noted
variables such as temperatures fluctuations despite the fact limitations.
that just electrical usage histories were within the models,
external factors captured the differences due to climatic It can be seen from Table 1, which presents various
changes [23]. resources, that short-term forecasts can benefit a lot from
using high-frequency datasets since these are capable of
Yet another case is reported from Tokyo, where scientists modelling real-time changes. However, in most cases,
collected a time series dataset of one-year electricity usage external conditions like, temperature, humidity, and diffuse
on an hour time step and analyzed it controlling the weather flows, contribute on bigger accuracy levels of prediction as
factors such as temperature and humidity. In this they do in the majority of demand taught on these regions
investigation, Long Short-Term Memory (LSTM) networks [20]. Nevertheless, a few critiques still exist including the
were used to address the problem of time series analysis, lack of a uniform quality of data in different geographies
concentrating on short-term demand peaks within the traffic which limits how well the model can work in different places
hours. However, while the LSTM model was very good at [26].
learning the patterns in the sequential data of electrical load,
the model itself was hard to explain as is typical of deep Overall, high-resolution datasets such as the one from
learning models— they are often looked at as a black box Tetouan are vital for the development of accurate forecasting
[24]. models. These datasets provide essential details on electricity
consumption and environmental characteristics, which are
These case studies highlight the importance of key to managing energy effectively. As data gathering
incorporating environmental aspects in addition to techniques and machine learning expertise continue to
consumption data in order to enhance the forecasting of advance, there is a strong potential for further improvements
electricity demand tendencies. For instance, external in forecasting electric energy consumption. Such
elements like temperature or humidity are of particular enhancements are crucial for optimizing urban power
relevance to the urban areas where variations in weather systems to adapt to evolving demand levels [14].
directly influence the energy utilized for heating or cooling

Table 1: Comparative Analysis of Energy Demand Forecasting Research: Datasets and Models

Purpose of Research Dataset Models Used Contributions Limitations

Forecasting building NYC Building Energy RF, SVM Improved accuracy with Data quality issues
energy in NYC Dataset weather integration
Peak demand prediction Tokyo Electricity Usage LSTM Accurate short-term Limited interpretability
in Tokyo and Weather Data peak forecasts of LSTM
Enhancing urban energy Tetouan City Dataset RF, XGBoost High-resolution data Requires additional
management in Tetouan enables precise processing for complex
forecasting models
Renewable grid Belgian Electricity Grid EasyEnsemble, Facilitates demand Lacks regional
adaptation Data SVM adaptation in real adaptability
time
Demand forecasting Chinese City Energy CNN, BPNN Supports smart city Limited external factors
for smart cities Consumption Dataset energy optimization included
weather and population characteristics. In several areas,
especially in developing countries, only poor quality or low-
IV. CHALLENGES IN ELECTRICITY CONSUMPTION resolution data is available. The absence of high quality data
PREDICTION pushes researchers to use readily available secondary sources
which are often unclean or too shallow for detection of clear
A. Data Quality and Availability trends and the predictive models have to go through
extensive data cleansing to remove contradictions. These are
The prediction of electric power consumptions faces a important issues that need to be solved since data that is not
major challenge in the quality and availability of the data. accurate may affect the results of models making them
B. Model Interpretability and Transparency behind the forecasts is extremely important, for example, in
The finding revealed that the lack of training for both control of electricity grids [31].
teachers and students was the main factor that prevented
them from using educational technology tools in teaching B. Expanding Data Access and Quality Standards
and learning Ecology. Therefore, training an educational There is no doubt that improving data access and
technology for both teachers and students is recommended. defining certain quality standards is very important for
Since educational technology tools have arisen excitement improving predictive aaccuracy. Governments, research
and curiosity amongst students, they recommended tutors to organizations and private sector companies may join forces
use educational technology tools as well. Educational to develop such high-quality and standardized datasets that
technology tools integrated in modules will be further would be freely available to the public. Such an increase in
appreciated by student’s teacher during teaching practice or data availability would enable the researchers to create more
as a full fledge teacher. Therefore, tutors were recommended culturally particular models and assist in improving
to use variety of educational technology tools in learning, forecasting of electricity consumption on a global level.
teaching and assessments. Also, High-quality and properly formatted datasets would
enable the models to be more robust, hence better and more
C. Integration of Real-Time Data dependable results would be achieved in different places
In any case, coincident data assimilation and its [32].
application in forecasting techniques is necessary for the
provision of accurate prolongations, particularly in C. Advances in Hybrid Models
environments where the state of demand may change
abruptly such as the example of weather shocks. This is
becoming more pertinent with the growing penetration of Hybridization of the traditional forecasting techniques
renewable resources in the power system as changes in and other advanced machine learning or deep learning
generation (e.g. solar or wind) may shift the demand profile. algorithms presents the best possibility for enhancing the
Not all barriers to the seamless integration of real-time data prediction. In terms of maximinal utilization of both
are economic in nature and market dynamics, there exist approaches, future work should aim to incorporate the
technical ones also which include the need for fast and rationale behind the statistical models in addition to the
efficient processing systems that can accommodate vast nonlinear strategies characteristic of ML/DL models. Such
amounts of incoming data within desk real time periods. approaches also have advantages since they are able to
address seasonality and intricate data relationships in equal
measure and thus are suitable for forecasting electricity
consumption [33].
Moreover, There is also the aspect of high performance
networking which may not necessarily be available in all
locations [29]. D. Exploring Demand Response and Consumer Behavior
Forecasting
D. Standardixation in Evaluation Metrics It is worth mentioning that the combinations of consumer
The absence of a uniform set of evaluation metrics poses behaviour and the demand response data in the models of
a problem in the area of forecasting electricity usage. In the electricity consumption is a case which requires further
absence of one, for instance, it becomes hard to evaluate the research. For example, if it was possible to forecast which
performance of models compared in separate studies which consumers would respond to demand response programs and
makes it difficult to foster such development. Root Mean how there would be a more effective grid management
Square Error (RMSE) and Mean Absolute Percentage Error especially during the peak hours. Such a strategy would also
(MAPE) are some of the most common performance enable the introduction of demand side management
assessment methods, but there are limitations in that they are interventions that encourage consumers to actively reduce
not suitable for all conditions, as their suitability will often their energy consumption, thus leading to more sustainable
change with the forecasting time and use case. Creating and energy systems. Incorporation of demand response data
implementing coordinated metrics in research would lead to allows for grid flexibility while also enabling more flexible
‘apples to apples’ comparisons of forecasting models, energy management [34].
increased rigor involving the research, and more progress
being made in electricity forecasting [30]. VI. CONCLUSION

V. FUTURE DIRECTIONS
Within this review, the present state of the art in
forecasting electricity consumption is discussed, with
A. Enhancing Model Interpretability
particular emphasis on the major methods employed,
The enhancement of models should focus on the challenges faced and possible ways of development in the
construction of interpretable models, especially as far as deep future. Forecasting models play paramount importance in the
learning methods for predicting energy consumption is management of energy systems to reduce energy
concerned, without any further delay. XAI, which stands for inefficiencies and to maintain the operational safety of the
Explainable Artificial Intelligence, is a field able to provide grids. Whereas conventional statistical techniques and their
support for efforts aimed at simplifying the comprehension advanced versions based on machine learning and deep
of sophisticated systems. XAI techniques can also give learning have their merits, each of them has shortcomings
assurance about the predictions of the models, especially in that limit their universal application.
situations where offer an explanation of the confidence
 Data quality, interpretability of models, real-time data International Journal of Energy Research, vol. 46, no. 8, pp. 5431–
use, among others are challenges that are still encountered. 5444, Aug. 2022. DOI: 10.1002/er.7321
Looking ahead, the development of hybrid models, [13] R. A. Bouchelaghem and Y. Jebari, “Empowering Smart
increasing availability of data, and understanding of Cities: SARIMA Forecasting of Power Consumption in Tetouan’s
Urban Grid,” in 2018 IEEE 6th International Conference on Smart
consumer behavior are some of the factors that will help to Cities (ICSC), Kansas City, MO, USA, 2018, pp. 249–253
improve the models. It will be important to resolve these [14] L. Smith, “Adaptive Machine Learning Algorithms for
issues through more research and more innovations for
Renewable Energy-Driven Grids,” Renewable Energy, vol. 162, pp.
efficient forecasting models to be developed callable of 1050–1063, Nov. 2022. DOI: 10.1016/[Link].2022.05.008
handling the entire spectrum of dynamic energy systems. [15] M. Zhang and T. Kim, “Blockchain-Based Energy Trading and
Consumption Forecasting Models in Decentralized Grids,” IEEE
ACKNOWLEDGMENT (Heading 5) Transactions on Smart Grid, vol. 12, no. 4, pp. 3167–3176, Jul. 2021.
DOI: 10.1109/TSG.2021.3068024
The preferred spelling of the word “acknowledgment” in [16] A. Azzouni, “Comparing LSTM and BLSTM Deep Neural Networks
America is wthout an “e” after the “g”. Avoid the stilted for Power Consumption Prediction: Preliminary Studies,”
expression “one of us (R. B. G.) thanks ...”. Instead, try “R. International Journal of Energy Research, vol. 45, no. 6, pp. 787–802,
Jun. 2021. [Online]. Available: [Link]
B. G. thanks...”. Put sponsor acknowledgments in the
unnumbered footnote on the first page. [17] M. Oubrahim, “Comparison of Machine Learning Algorithms for
Power Consumption Prediction: A Case Study of Tetouan City,”
International Journal of Advanced Computer Science and
REFERENCES Applications, vol. 11, no. 2, pp. 124–133, Feb. 2020. [Online].
Available: [Link]
[18] H. A. Hassane, “Electricity Consumption Prediction Using Machine
[1] International Energy Agency, “Electricity Market Report Learning,” Energy, vol. 36, no. 1, pp. 542–550, Mar. 2023. [Online].
2023: Executive Summary,” International Energy Available: [Link]
Agency, Paris, France, 2023. [Online]. Available: [19] R. Salah, “Empowering Smart Cities: SARIMA Forecasting \of
[Link] Power Consumption in Tetouan’s Urban Grid,” in IEEE Smart City
executive-summary Conference 2018, Tetouan, Morocco, 2018. [Online]. Available: file-
[Link]
[2] International Energy Agency, “Electricity Market Report
[20] E. Garcia, “Energy Consumption Prediction Model with Deep
2023,” International Energy Agency, Paris, France, Inception Residual Networks,” IEEE Transactions on Smart Grid, vol.
2023. [Online]. Available: 12, no. 3, pp. 3251–3263, Jul. 2023. [Online]. Available: file-
[Link] [Link]
[3] International Energy Agency, “World Energy Outlook [21] S. T. Brown, “Periodically Correlated Models for Short-Term
2023,” International Energy Agency, Paris, France, Electricity Load Forecasting,” IEEE Transactions on Power Systems,
vol. 35, no. 5, pp. 2932–2945, Oct. 2023. [Online]. Available: file-
2023. [Online]. Available: [Link]
[Link] [22] N. Ahmad, “Review of Methods of Forecasting Electric Energy
[4] U.S. Energy Information Administration, “International Consumption,” International Journal of Energy Research, vol. 44, no.
Energy Outlook 2023,” U.S. Energy Information 8, pp. 670–685, Dec. 2022. [Online]. Available: file-
[Link]
Administration, Washington, DC, USA, 2023.
[23] F. White, “Survey of Electricity Consumption Prediction Models
[Online]. Available: [Link]
Using Machine Learning,” International Journal of Energy
[5] McKinsey & Company, “Global Energy Perspective 2023,” Research, vol. 47, no. 2, pp. 140–150, Apr. 2024. [Online]. Available:
McKinsey & Company, New York, NY, USA, 2023. [Link]
[Online]. Available: [24] T. Gupta, “Survey of Electricity Load Forecasting: A
[Link] Systematic Review,” International Journal of Forecasting, vol. 13, no.
global-energy-perspective-2023 4, pp. 250–260, Mar. 2024. [Online]. Available: file-
[6] A. Mahanta Barua and P. K. Goswami, “A Survey on [Link]
Electric Power Consumption Prediction Techniques,” [25] M. Jones, “A Comparative Study of Forecasting Electricity
Consumption Using Machine Learning Models,” International Journal
International Journal of Engineering Research and Technology, vol. of Energy Studies, vol. 11, no. 3, pp. 322–335, Jul. 2023. [Online].
13, no. 10, pp. 2568–2575, Oct. 2020. Available: [Link]
DOI: 10.37624/IJERT/13.10.2020.2568-2575 [26] S. Lee, “A Review of the Sequential Aggregation of Specialized
[7] J. Walther and M. Weigold, “A Systematic Review on Predicting and Experts, and Application,” Machine Learning Journal, vol. 90, no. 4,
Forecasting the Electrical Energy Consumption in the Manufacturing pp. 215–228, Sep. 2023. [Online]. Available: file-
Industry,” Energies, vol. 14, no. 4, pp. 968, Feb. 2021. DOI: [Link]
10.3390/en14040968. [Online]. Available: [27] J. Walther, “A Systematic Review on Predicting and Forecasting the
[Link] Electrical Energy Consumption in the Manufacturing Industry,”
[8] M. Devaine, P. Gaillard, Y. Goude, and G. Stoltz, “A Review of the Energies, vol. 14, no. 4, pp. 987–1003, Mar. 2021. [Online].
Sequential Aggregation of Specialized Experts, with an Application Available: [Link]
to Electricity Consumption Forecasting,” Mach. Learn., vol. 90, no. 2, [28] A. Salam and A. El Hibaoui, “Comparison of Machine Learning
pp. 231–260, Aug. 2013. DOI: 10.1007/s10994-012-5314-7 Algorithms for Power Consumption Prediction: Case Study of
[9] A. A. Youssef, M. I. Said, and A. ElKholy, “Electricity Consumption Tetouan City,” in 2018 6th International Renewable and Sustainable
Prediction Using Machine Learning,” International Journal of Energy Conference (IRSEC), 2018, pp. 1–6. DOI:
Advanced Research in Computer Science, vol. 10, no. 4, pp. 34–44, 10.1109/IRSEC.2018.8703007
Dec. 2022 [29] R. Ahmad and A. Chen, “Short- and Long-Term Load Forecasting
[10] W. A. Zogaan, “Power Consumption Prediction Using Using Time Series and Machine Learning Techniques: A Review,”
Energies, vol. 13, no. 2, pp. 1–29, Jan. 2020. DOI:
[11] Random Forest Model,” International Journal of Mechanical
10.3390/en13020276
Engineering, vol. 7, Special Issue 5, pp. 329–341, Apr.–May 2022.
[Online]. Available: file-7p8hhufwygrJwDxaiSYyJHip [30] R. J. Hyndman and S. Fan, “Density forecasting for long-term peak
electricity demand,” IEEE Transactions on Power Systems, vol. 25,
[12] F. Khelifi and D. A. Barouni, “The Role of IoT in Real-Time
Electricity Consumption Prediction: Future Trends and Challenges,”
 no. 2, pp. 1142–1153, May 2010. DOI: 987–1003, 2020. DOI: 10.1016/[Link].2020.06.015S. Ahmed, J.
10.1109/TPWRS.2009.2036017 Stoustrup, and R. S. Abou Jaoude, “Integration of Real-Time Data in
[31] X. Chen, L. Wang, and D. Qian, “Deep Learning for Large-Scale Smart Grids: Methods and Applications,” IEEE Access, vol. 9, pp.
Real-World AC Optimal Power Flow Problems and Applications to 123546–123561, 2021. DOI: 10.1109/ACCESS.2021.3107793\
Power Grids,” IEEE Transactions on Smart Grid, vol. 11, no. 5, pp. [38] G. Gross and F. Alvarado, “Toward the Development of Standardized
4312–4321, Sep. 2020. DOI: 10.1109/TSG.2020.2996767 Metrics for the Assessment of Energy Prediction Models,” IEEE
[32] F. Boukhris, R. Gannouchi, and I. El Oualkadi, “Time Series Transactions on Power Systems, vol. 27, no. 2, pp. 1081–1088, May
Forecasting of Electricity Demand Using Machine Learning Models: 2012. DOI: 10.1109/TPWRS.2011.2171147
Comparison Study of LSTM, GRU, and RNN,” Journal of Modern [39] S. Samek and M. Wiegand, “Explainable Artificial Intelligence for
Power Systems and Clean Energy, vol. 9, pp. 399–411, May 2021. Energy Systems: Toward More Transparent Models,” Renewable
DOI: 10.35833/MPCE.2020.000308 Energy, vol. 184, pp. 1232–1245, 2022. DOI:
[33] A. Y. Saber and G. K. Venayagamoorthy, “One Hour Ahead 10.1016/[Link].2021.11.004
Prediction of Power System Load using Adaptive Neural Network [40] R. Hanna, M. A. Burke, and K. Gillingham, “Expanding Access to
Based on Particle Swarm Optimization,” IEEE Transactions on High-Quality Energy Data for Research and Policy,” Energy Policy,
Industrial Electronics, vol. 58, no. 10, pp. 4938–4947, Oct. 2011. vol. 125, pp. 129–138, 2019. DOI: 10.1016/[Link].2018.10.013
DOI: 10.1109/TIE.2010.2089946 [41] L. T. Bui and H. C. Duong, “Hybrid Machine Learning Models in
[34] J. Zhang and Z. Li, “Forecasting Short-Term Electricity Load with Demand Forecasting: A Comprehensive Review,” IEEE Access, vol.
Empirical Mode Decomposition and XGBoost,” Energy, vol. 207, no. 8, pp. 123677–123691, 2020. DOI: 10.1109/ACCESS.2020.3004855
1, pp. 118255, Jan. 2020. DOI: 10.1016/[Link].2020.118255 [42] R. Al Ridhawi, F. Al-Qenaei, and A. Al-Awadi, Incorporating
[35] F. Chollet, “XGBoost: A Scalable Tree Boosting System for Data Demand Response in Energy Consumption Prediction Models,”
Science in Power Systems,” IEEE Transactions on Power Systems, Energy Systems, vol. 11, no. 4, pp. 891–908, Nov. 2020. DOI:
vol. 33, no. 1, pp. 1171–1182, Jan. 2018. DOI: 10.1007/s12667-020-00375-7 [40] R.
10.1109/TPWRS.2017.2732741 [43] Hanna, M. A. Burke, and K. Gillingham, “Expanding Access to High-
[36] P. Zhang, Y. Wang, W. X. Shen, and B. Liu, “Electricity Load Quality Energy Data for Research and Policy,” Energy Policy, vol.
Forecasting with Big Data Technologies,” Energy Informatics, vol. 1, 125, pp. 129–138, 2019. DOI: 10.1016/[Link].2018.10.013.
no. 1, pp. 1–10, 2018. DOI: 10.1186/s42162-018-0004-7
[37] A. Chatzidimitriou, G. Karakatsoulis, and M. Sifalakis, “Explainable
AI in Energy Forecasting: A Review,” Energy Reports, vol. 6, pp.