1. INTRODUCTION

The growing need to have sustainable power and the incorporation of renewable energy sources have prompted the shift towards conventional systems of power to Smart Grids (SGs) [1]. There are also two-way communication and the ability to self-heal and dynamically balance loads, which are essential to the performance of modern energy systems unlike traditional grids, and SGs can do both (Jha 2015, p. 2). In this regard, the concept of Artificial Intelligence (AI) has become a transformational technology, which can be used to achieve predictive modeling, adaptation control, and optimization in energy management [3].

Proper load prediction is among the basic needs of SG operation. Traditional statistical models are usually incapable of identifying the nonlinear nature of energy consumption. Random Forests and Deep Learning are AI-based approaches that have shown better results in recognizing the seasonal changes and sophisticated consumption patterns [4, 5]. As an example, it has been noted by recent research that ensemble learning can significantly enhance the precision of short-term demand forecasting in grids with significant renewable contents [6].

The renewable energy resources (RERs) are challenging to integrate, especially solar and wind, because they are intermittent in nature [7]. The solar irradiance and wind patterns have been forecasted using AI with enhanced accuracy to enable the grid operators to make more optimal plans and less curtailment [8]. Moreover, it has been suggested that hybrid models using physical simulations and AI methods will be more effective in harvesting renewable integration [9]. These methods would guarantee that stability of the grid is achieved whilst achieving sustainability goals [10].

Another area of great significance that AI has been implemented is grid optimization. Resource allocation and energy distribution in SGs at the intersection of optimization algorithms, including reinforcement learning, linear programming, and metaheuristic methods, are underway [11]. These techniques can be used to allocate resources in real-time, especially to balance the loads between the producers, the consumers and the storage units [12]. Recent studies indicate that AI-based optimization is capable of minimizing energy waste and enhancing grid resilience throughout very poor operating conditions [13].

Other key elements of reliable SG operations are real-time monitoring and anomaly detection. Through sensor networks and Internet of Things (IoT) devices, AI systems would be able to identify anomalies, equipment malfunctions, or cyberattacks at the initial stages [14]. Isolation forest and Autoencoders are examples of machine learning models used to identify unusual demand peaks or grid operations deviations [15]. Proactive interventions can be made through early detection, and this is much more reliable [16].

The importance of energy storage systems also lies in facilitating the process of matching demands and supplies, particularly when the percentage of renewable energy is large [17]. Predictive control algorithms that are based on AI have been created in order to streamline storage utilization, improve battery duration, and lower operating expenses [18]. Additionally, the inclusion of electric vehicles (EVs) as mobile storage devices introduces new opportunities and challenges and necessitates AI-based scheduling and demand response plans [19].

Irrespective of such developments, the adaptation of AI to SGs is still faced with challenges. Compromises like privacy of data, interoperability, and cybersecurity are of great concern. Massive implementation should have safe communication systems, as well as privacy-sensitive AI algorithms [20]. Moreover, it is still a gap in the research to achieve scalability of models in heterogeneous grid environments.

2. LITERATURE REVIEW

The recent developments of smart grids (SGs) have shown how the Artificial Intelligence (AI) helps to optimize the smart grids, predict, find anomalies and integrate the renewable sources of energy. The most suitable models are recurrent and attention based models, which predict energy use [22], personalised federated learning model can predict a smart meter load [23], Taik and Cherkaoui integrate edge computing and federated learning to predict a smart meter load with maximum accuracy and privacy [24].

In addition, new AI-specific load forecasting methods have already been tried in terms of resilience to changing conditions [25]. Besides demand, AI-integrated forward chain projections of renewable generation projections have been suggested recently, demonstrating improved solar and wind combination [26], or renewable resource combination issues in SGs [27]. The fundamental principle of the SG performance is optimization and the methodical examination on the use of heuristic and metaheuristic algorithms in energy management [28-30], sensing and understanding [31] and demonstration of reinforcement learning methods in energy scheduling [31-32], have been introduced. Uncertainty robust operation optimization uncertainty operation in organic RGPT [32], Tier-Based IA-CoO of a grid with an optimized and secure operation under uncertainty [33].

Reliability also relies on the anomaly detection and cyber security, and ML-based attack detection in grid streams [34], anomaly detection specific to industrial control systems [35], and the survey of cyber threats and defense mechanisms in SGs [36]. In line with this, emphasized big data analytics and ML applications to SG anomaly response [37]. Storage of energy and demand flexibility also play a crucial part, since optimal scheduling of battery systems has been studied in literature [38], hybrid microgrids with storage and EV have been studied [39], Demand response strategies have been reviewed widely, focusing on the coordination of user consumption with dynamic supply [40], AI-coordinated EVs, storage, and the grid [41]. Nevertheless, issues remain, as the scalability of AI solutions in heterogeneous SG settings is still being explored [42], data privacy issues are still getting larger [43], and explainable AI is becoming a well-known demand to the operational trust [44]. Taken altogether, these investigations prove that although AI has made impressive progress in terms of SG forecasting, optimization, monitoring, and control, future studies should focus on interoperability, trustworthiness, and scalability of deployment to realize the full potential of a sustainable and intelligent energy distribution.

3. METHODOLOGY

Data Generation and Preprocessing

The synthetic dataset was created as a three-year (2022-2024) dataset (daily resolution) in order to be able to model energy demand, renewable supply (solar, wind), grid stability, and weather (temperature, cloud cover). The model of data generation was supposed to provide a realistic variation in both energy demand and generation of renewable energy with consideration of both seasonal and cyclic factors.

Energy Demand Modeling

A combination of a baseline trend, seasonal variations (annual and weekly cycles), and Gaussian noise were used to model the daily energy demand D(t)D(t)D(t): accounted by the randomness of demand. The following is the definition of demand model:

 …... (1)

In eq(1) is the energy demand at time t. α, β, and γ are coefficients that control the baseline demand, annual seasonal variation, and weekly cycles, respectively. δ represents a linear trend, and is Gaussian noise, ensuring that demand fluctuates with random variation.

Solar Generation Model: Solar energy generation was modeled as a sinusoidal function with an annual cycle, modulated by wind conditions

 ……………… (2)

Where equation (2) is the energy generated by solar power at time t. η is the solar generation efficiency factor. ϕ is the phase shift representing the solar energy cycle's start, and W(t)represents the weather influence on solar generation.

Wind Generation Model: Similarly, wind energy generation was modelled as:

…………..(3)

Eq(3) is the energy generated by wind at time t. μ and κ are wind generation coefficients, and is the wind speed factor at time t. λ is the phase shift representing the wind energy cycle's start.

Total Supply: The total energy supply is the sum of solar, wind, and grid supply  in eq(4).

 ……………… (4)

Energy Balance: The difference between demand and supply is referred to as the energy balance which is expressed as follows in equation (5):

 ………………………… (5)

Normalization, feature scaling, and lagged features were used as preprocessing. Linear interpolation of missing values was performed.

Table 1: Summary Statistics of Generated Data

Feature Engineering

Temporal dependencies were utilized by generating lagged features of demand, solar generation and wind generation. Demand lagged features were formed as an example:

 …(6)

Where , and  are the demand, solar, and wind features, respectively in this eq(6), and and represent temperature and cloud cover at time t.

Normalization Function: Data values were scaled to a range between 0 and 1 to facilitate the model training and prevent dominance of features with higher numerical values.

 ……………………..… (7)

Here eq (7)  is the original feature, and and are the minimum and maximum values of the feature.

Feature Engineering

The relationships among different factors influencing the energy demand and supply were learnt through a feature correlation analysis.

Table 2: Feature Correlation with Demand

Table 2 Description:

This table is used to show the values of the correlation between the various features (demand, solar, wind, temperature, and cloud cover) and the target feature, which in this case is demand. It is noteworthy that the demand is strictly linked with lagged demand characteristics, which suggests that the previous demand has a powerful impact on the upcoming demands in energy. Also, the weather conditions like temperature are positively correlated to the demand and cloud cover has negative effects on solar energy production.

Demand Forecasting Model

The random Forest Regressor (RFR) model was used to generate demand forecasting, where 80% of the data was used as the training set, and the remaining 20% was used as a test set. The model tells the energy demand based on the feature vector generated in the process of feature engineering. Prediction Model:

 ……………….……... (8)

Equation (8) here is the predicted demand, T is the total number of time steps, and is the forecasted value at time t.

Evaluation Metrics: The following metrics were used to evaluate the performance of the model:

Mean Squared Error (MSE):

 ……………..……… (9)

Coefficient of Determination (R²):

 ………………..(10)

Root Mean Squared Error (RMSE):

 ……………….(11)

Mean Absolute Percentage Error (MAPE):

 …………………. (12)

The RFR achieved R² = 0.80 and MSE = 7.25, validating predictive capability.

Smart Grid Network Simulation

The grid network was modeled as a small-world network (Watts-Strogatz) of 12 nodes. The nodes were an indicator of a consumer, producer, or a storage.

Allocation Rule:

 …………………..… (13)

Where is the demand at node n and S is the total available supply.

Energy Balance Constraint:

 ……… (14)

This constraint ensures that the total allocation from consumer nodes (C) and storage nodes (S) does not exceed the total available grid supply .

Resilience metric:

 ………………………………………..… (15)

V denotes the total number of nodes and V a denotes the numbers of nodes that are non-failing. The resilience measure measures the capability of the network to keep operating as a result of failures. Minimization of energy consumption:

The energy allocation optimization process is described in algorithm 1 and it will allocate the energy to the consumer nodes and subsequently to the storage nodes, and dynamically adjusts to the demand and supply at real time.

Input: Graph G(V,E), total supply S

Initialize allocations = 0

For each consumer node c ∈ V:

allocate = min(demand(c), S)

S = S - allocate For each storage node s ∈ V:

allocate = min(capacity(s), S)

S = S - allocate

Output: allocations

Real-Time Simulation

A real time simulation of 14 days was done and it combined the forecasted demand together with stochastic generation of the renewable. Adaptive Control Law:

 ………… (16)

Battery Storage Dynamics:

 ……………………………….(17)

Consumer Satisfaction Index (CSI):

 ………………………… (18)

Renewable Penetration Rate (RPR):

 ……………………….(19)

Sustainability Index (SI):

 …………………… (20)

Table 3: Real-Time Simulation Metrics

Table 3 this table shows the major performance indicators of the real-time simulation such as the forecast accuracy (R 2), the percentage of days with excess energy and the percentage of days with renewable energy in the supply. It also gives the amount of anomalies it detected in demand showing the capability of the model to indicate irregularities.

4. RESULTS

Forecasting Performance

Mean Squared Error (MSE) and R 2 were considered to be the important regression measures used to evaluate the performance of the energy demand forecasting model. The MSE was found to be 7.25 and the R 2 was found to be 0.80 and this indicates that the model was able to explain 80 percent of the variation in demand of energy and this means that the model has a strong fit to predict energy demand. These findings imply that the predictive model is efficient to predict the future energy requirements taking into consideration seasonal and non-seasonal changes in demand.

Figure 1: Analysis of the energy demand and supply.

The figure 1 concentrates on the interactions of energy demand and supply and specifically the imbalance between supply and demand and identification of irregularities. (a) Energy Demand and Supply Over Time This figure is a comparison between the demand of energy (in blue) and the total energy supply (in green) over time. It shows the changes in demand as well as supply and the red shaded areas demonstrate the times when the demand exceeded the supply and this indicated energy shortages. The data highlights the need to have predictive models and an optimized grid that will be able to balance such differences in such a way that the energy will be reliable. (b) Distribution of Demand and Supply a histogram of distribution of energy demand (in blue) and the total supply (in green). The distribution gives us a chance to see the scope of values of demand and supply, but with more focus on how demand and supply are not at par with each other. This value clearly proves that there exist considerable periods when the supply is not enough to satisfy the demand, which stresses the importance of the sophisticated predictions and optimization to avoid possible variations in power supply. Please, note: (c) AI-Detected Anomalies in Demand Data This time series plot shows the energy demand over time (in blue), where the anomalies identified by the AI model are marked by red dots. The red points are a sign of the existence of outliers or non-general spikes on the demand data, which demonstrates the capability of the model to identify the presence of abnormal consumption patterns. Such anomalies are crucial in the identification of any possible problems like grid faults or sudden load surges that can be dealt with early enough.

Figure 2: Renewable Energy Integration

In the figures 2 below, the incorporation of renewable energy sources, that is, solar and wind power into the grid and the sustainability of these sources to the energy requirements is explored. (a) Renewable Energy Generation The figure depicts the total energy produced during renewable energy generation (green line) and versus the solar (orange line) and wind (blue line) energy. It emphasizes the nature of fluctuation of renewable energy sources especially solar and wind that vary significantly during the year. Nevertheless, regardless of these variabilities, the overall renewable energy production is very important in ensuring that the grid is sustainable and it does not overly depend on fossil fuels. (b) Sustainability Trends: Renewable Contribution & Surplus Days Plot illustrating rolling average of the ratio of renewable energy in the grid supply (green curve) and the ratio of the surplus days in the grid (blue curve). The rolling averages show how the renewable sources will be effective towards maintaining a steady supply of energy during the simulation period. The excess days are seen when the energy supply is greater than the demand, and it can be stated that the grid was efficient enough, and the work was performed mostly with the help of renewable sources without the depletion of resources.

Figure 3 is used to show the optimization of energy in the grid nodes, which are storage, consumer, and production nodes. Figure 3(a) Energy Allocation to storage nodes over time this plot is an indication of the way energy has been distributed across the storage nodes (Storage 4 and Storage 10) over time. It is interesting to note that no storage node was allotted any energy over the simulation period. This might be because of certain set ups or assumptions in the model like the storage capacity is either full or redundant within the conditions of the simulation. Figure 3(b) Adaptive Allocation to Consumer Nodes Over Time The figure shows how the energy is dynamically allocated to the various consumer nodes (colored lines) during the period of simulation. The algorithm modifies the energy distribution by the needs of every node. When there is high demand, the algorithm will allocate more energy to key consumer nodes and this is why in all cases the most important consumers will be assigned the priority. This dynamic balancing is essential in ensuring there is a balance between supply and demand at all the grid nodes.

The next figure 4 is devoted to the forecasting model performance regarding the future energy demand. (a) AI Forecast vs. Actual Energy Demand This chart shows the comparison of the forecasted energy demand (magenta dashed line) versus actual energy demand (blue solid line) with time. It shows that the AI forecasting model is very accurate that the model prediction of the demand is very close to the actual demand. The few differences between the two curves can be taken to denote regions where the forecasting process may be further enhanced, and on balance, the model performs quite well. (b) Predicted vs. Simulated Actual Demand AT comparison between the actual demand (magenta dashed line) predicted (magenta dashed line) and that of the actual demand (green solid line) over time. This value can be used to determine the effectiveness of the AI model in modeling the dynamics of demand as opposed to real data. The minor discrepancies indicate that the model can be used to reproduce demand patterns, and also indicate that there is sometimes a mismatch which can be refined.

Figure 3: Grid Optimization and Allocation

Figure 4: Forecasting Performance

Figure 5: Weather and Environmental Impact

This Figure 5 discusses the impact of weather conditions on renewable energy production and on the overall grid performance especially temperature and cloud cover. (a) The result of this plot is the temperature (red line) versus the cloud cover (grey line) with respect to time. One of the critical factors in the production of solar and wind energy is the temperature. The figure equally demonstrates the cloud cover effect on the production of solar energy with an increase in cloud cover decreasing the production and production of solar energy and it is important to note that there should be adaptive energy management plans that take into consideration the effect of weather changes. (b) Feature Correlation Heatmap This heatmap shows the association of the different features such as the energy demand, solar production, wind energy, grid supply, temperature and cloud cover. Darker colors represent the existence of high positive and negative correlations and, therefore, assist in detecting the relationships between weather factors and energy generation/consumption. As an example, the inverse relationship between solar production and cloud cover indicates that the more a cloudy cover, the lower the production of solar energy which is an important consideration in the optimization of renewable energy production.

Figure 6: Grid Network Representation

The Figure 6 gives any understanding of the structural configuration of the grid and the distribution of energy among nodes. (a) Smart Grid Network a network diagram depicting how the smart grid is set up, where the consumer (blue), producer (green) and storage nodes (orange) are placed. This visualization aids in the comprehension of the energy flow between the various nodes and this is essential in comprehending the way the grid is set and how to provide maximum energy to each and every unit. (b) Feature Correlation Heatmap Another heatmap of feature correlation. This value is critical in learning the way in which energy distribution may be optimized among the various grid nodes depending on the interrelationships between demand, supply and other environmental aspects.

5. DISCUSSION

Good indications of the success of the proposed AI-based smart grid system to address the challenges of forecasting, allocation, and adaptive control are found in the findings of this paper to implement the sustainable energy distribution. The results are critically discussed in this debate and compared with the previous literature and the implications raised on the basis of future research and application of the grid in practice.

Validity of Forecast and Strength of Model

The forecasting experiment revealed that the ensemble learning was highly correct with R 2 = 0.80 and MSE = 7.25. The following values ascertain the fact that the model is viable in capturing the complex seasonality, long term patterns of demands, and short term fluctuations that characterize the modern energy systems. It is important to note that the demand lags were the most effective predictors but other contextual factors were found in renewable generation and weather-related inputs. This confirms that combos of features improve prediction accuracy without affecting the interpretability. This interpretability is key to operator trust in functional smart grid systems, and is congruent with the overall opinion that the explainable AI practices should be accompanied with accuracy enhancement.

Energy Allocation and Grid Equity

The simulations of the energy optimization supported the notion that the allocated algorithm developed will ensure the adequacy of supply under the circumstances of the varying demand and the renewable energy. There were generally sufficient allocations about and storage units were dynamically used to either absorb excess energy, or to provide back where required. Though a further inquiry reveals that the equity guarantee between the consumers was not absolute since higher allocation was always made on some nodes as opposed to others. It is a drawback of greedy allocation strategies; they are helpful to prevent shortages, but they do not always address the question of fairness. In practice, a high motivation will arise to integrate the fairness-based optimization techniques so as to give equal accessibility especially in a heterogeneous community in which its communities have different energy needs.

Sustainability and Real-Time Adaptive Control

The real time simulation determined that the forecasting model would be successfully incorporated in the operational decision making. The demand across all training and test combinations generalized the forecasted demand which was highly consistent with the simulated consumption. On sustainability, the mean renewable penetration was 46.5 which indicated that it was adequate all through the simulation horizon. The combination of moderateness and renewability is a major achievement, as it implies that AI-based grids are capable of meeting the demand and simultaneously help to achieve sustainability. The other problem that can be solved by the smart forecasting and smart allocation is the variability, which is also signified by the fact that the renewable contributions have been steady.

Anomaly Detection and operational Resilience

Approximately 3 percent of the daily demand points were detected as abnormal by the anomaly detection committee which is usually associated to a significant shift in demand. The anomalies are operationally relevant, and they are conditions, which are most likely to lead to the inaccuracy of forecasting and productivity of allocating. Prevention control measures provide an early warning of such events, which will be preemptive storage charges, activation of rapid demand response, or temporary utilization of reserve capacity. Through the incorporation of the concept of anomaly detection into the grid real-time control, one will enhance resilience to ensure that AI-controlled grids will not be subjected to too high variations and other unforeseen operational issues.

Connotations of Extended Smart Grids

The integrated framework has a number of larger implications. First of all, the history of incorporating forecasting in the allocation decision-making process is the way predictive analytics can influence positively the operations planning process bridging the gap between planning and real-time control. Second, it emphasizes the necessity to build a responsive to concerns of fairness: augmenting adequacy is not in itself equitable access, the allocation policy must vary to bring about a balance between technical and social objectives. Finally, but not the least, the element of anomaly detection demonstrates the fact that AI systems in infrastructure critical areas cannot be reduced to optimization and prediction as well as incorporating resilience and reliability.

Limitations and Opportunities

Even though the current study has demonstrated good results, several limitations are acknowledged. The synthetic data can be used in testing, however, it should be verified with the actual grid data to verify that the model can be generalized. Though it has an efficient allocation mechanism, which is adequate, it remains poor in its fairness and long-term optimization in its storage. Similarly on the same note, the anomaly detection model can provide first line resilience but it must be provided with context sensitive filtering so that it can distinguish normal variability and significant threats. This can be used to introduce opportunities in potential future studies, including reinforcement learning-based allocation, multi-agent systems to allocate distributed resources, and explainable anomaly detection models that combine statistical and domain-based knowledge.

Agility with Present Research

The article is appropriate in the new literature that emphasizes the significance of the combination of forecasting, optimization, and resilience processes to grids operated by AI. It demonstrates that machine learning can render the balance between adequacy, sustainability, and fairness feasible and offer a reasonable framework to adhere to during the implementation of pilots in the future. The forecasting, allocation and anomaly detection are placed in the same pipeline, and the study reveals the potential of AI not just as a tool of analysis, but also as an internal component of the systems of energy management of the future.

6. CONCLUSION

The present research presented an AI-based smart grid architecture that seeks to become sustainable through the promotion of energy distribution by forecasting, real-time adaptive control, and anomaly detection. The analysis integrated the synthetic data generation, feature engineering, ensemble-based demand forecasting, network simulation of smart grid and resilience analysis into a single pipeline. The results indicate that artificial intelligence can play a critical role in resolving the multi-dimensionality of the modern power systems.

The forecasting model used was of great predictive power and its R 2 was 0.80 and the mean squared error was 7.25. These findings confirm the applicability of the ensemble learning algorithm in the procedure of reproducing seasonal changes along with stochastic variations in energy demand. Specifically, the contextual accuracy could be improved by including renewable and weather functionalities that could be utilized to advance climate-sensitive signals in the forecasting pipelines. This observation brings to fore that predictable and data-driven models must be plausible in use in critical infrastructure.

The experiments of the network allocation showed that the adequacy was never compromised and that this ensured that the consumer demand was maintained and the storage units were working intensely in an attempt to harmonize the supply and demand. The allocations also distributed however helped to clarify the issues of equity in the sense that there was imbalanced distribution of both the consumers and the storage nodes. This outcome can be used to emphasize the importance of fairness-conscious algorithms in the design of the future smart grid. Adequacy is fundamentally a primary need but fair access to energy is also fundamental to state that the society believes in smart infrastructure.

The proposed system was demonstrated to be stable during dynamic operating conditions as reflected in the real-time simulations. The contribution of the renewable was 46.5 percent on average throughout the study period and the sufficiency of supply made was 100 percent. Such results prove that sustainability and reliability are both possible in the event of the integration of AI-driven forecasting and adaptive allocation systems. Giving sufficient penetration with ensuring that it is renewable, the framework is intended to address one of the most persistent issues in the energy management balancing between the environmental agenda and the stability of operations.

Anomaly detection was also significant so as to enhance the resilience of operations. The anomalous demand points detectable were approximately 3 percent and were mostly followed by the unexpected high increases or decreases. It is an early-warning system that can be applied as a buffer against interruption to ensure that the operators can have the capacity to bring in demand response, exploit storage capacity, or access backup grid resources at an earlier time. The resilience mechanisms are particularly important during the periods when the smart grids are becoming increasingly dependent on renewable electricity that is increasingly becoming volatile, and the decentralized energy sources.

This study also has some general implication to the design of the future energy systems beyond the technical contributions. It demonstrates that intelligent forecasting and adaptive control is viable to be applied within one structure and bridging between planning and management is eliminated. It also underlines that the concept of fairness and resilience needs to be the priority besides the concepts of adequacy and sustainability, in order to make the transition to AI-based grids relevant to the technical performance and societal needs.

Even though the outcomes are promising, there remain areas of weaknesses. The implication of resting on artificial data is that there is need to compare it to more realistic grid data to ascertain generalizability. Besides, the greedy allocation mechanism, though effective in the adequacy, is not ideal as far as the storage performance of a long-term scale and fairness measures are concerned. The future studies should include reinforcement learning and multi-agent coordination and fairness-sensitive optimization to extend the scope of application of this framework. Besides, explainable anomaly detection techniques may increase the trust in the operators as they can distinguish the threat level of critical variability and benign variability.

REFERENCES

1. Fang, X., Misra, S., Xue, G., & Yang, D. (2011). Smart grid—The new and improved power grid: A survey. IEEE communications surveys & tutorials, 14(4), 944-980.
2. Mohsenian-Rad, A. H., & Leon-Garcia, A. (2010). Optimal residential load control with price prediction in real-time electricity pricing environments. IEEE transactions on Smart Grid, 1(2), 120-133.
3. Etesami, S. R., Saad, W., Mandayam, N. B., & Poor, H. V. (2018). Stochastic games for the smart grid energy management with prospect prosumers. IEEE Transactions on Automatic Control, 63(8), 2327-2342.
4. Habbak, H., Mahmoud, M., Metwally, K., Fouda, M. M., & Ibrahem, M. I. (2023). Load forecasting techniques and their applications in smart grids. Energies, 16(3), 1480.
5. El-Telbany, M., & El-Karmi, F. (2008). Short-term forecasting of Jordanian electricity demand using particle swarm optimization. Electric power systems research, 78(3), 425-433.
6. Contreras, J., Espinola, R., Nogales, F. J., & Conejo, A. J. (2003). ARIMA models to predict next-day electricity prices. IEEE transactions on power systems, 18(3), 1014-1020.
7. El-Hawary, M. E. (2014). The smart grid—state-of-the-art and future trends. Electric Power Components and Systems, 42(3-4), 239-250.
8. Raza, M. Q., & Khosravi, A. (2015). A review on artificial intelligence based load demand forecasting techniques for smart grid and buildings. Renewable and Sustainable Energy Reviews, 50, 1352-1372.
9. Osório, G. J., Lotfi, M., Shafie-Khah, M., Campos, V. M., & Catalão, J. P. (2018). Hybrid forecasting model for short-term electricity market prices with renewable integration. Sustainability, 11(1), 57.
10. He, C., Dai, C., Wu, L., & Liu, T. (2018). Robust network hardening strategy for enhancing resilience of integrated electricity and natural gas distribution systems against natural disasters. IEEE Transactions on Power Systems, 33(5), 5787-5798.
11. Zheng, W., Liu, H., & Sun, F. (2020). Lifelong visual-tactile cross-modal learning for robotic material perception. IEEE transactions on neural networks and learning systems, 32(3), 1192-1203.
12. Mohsenian-Rad, A. H., Wong, V. W., Jatskevich, J., Schober, R., & Leon-Garcia, A. (2010). Autonomous demand-side management based on game-theoretic energy consumption scheduling for the future smart grid. IEEE transactions on Smart Grid, 1(3), 320-331.
13. Zahraoui, Y., Korõtko, T., Rosin, A., Mekhilef, S., Seyedmahmoudian, M., Stojcevski, A., & Alhamrouni, I. (2024). AI applications to enhance resilience in power systems and microgrids—A review. Sustainability, 16(12), 4959.
14. Ozay, M., Esnaola, I., Vural, F. T. Y., Kulkarni, S. R., & Poor, H. V. (2015). Machine learning methods for attack detection in the smart grid. IEEE transactions on neural networks and learning systems, 27(8), 1773-1786.
15. Dehlaghi-Ghadim, A., Moghadam, M. H., Balador, A., & Hansson, H. (2023). Anomaly detection dataset for industrial control systems. IEEE Access, 11, 107982-107996.
16. Nguyen, T. N., Liu, B. H., Nguyen, N. P., & Chou, J. T. (2020, June). Cyber security of smart grid: attacks and defenses. In ICC 2020-2020 IEEE International Conference on Communications (ICC)(pp. 1-6). IEEE.
17. Ho, W. S., Macchietto, S., Lim, J. S., Hashim, H., Muis, Z. A., & Liu, W. H. (2016). Optimal scheduling of energy storage for renewable energy distributed energy generation system. Renewable and Sustainable Energy Reviews, 58, 1100-1107.
18. Salkuti, S. R. (2019). Optimal operation of microgrid considering renewable energy sources, electric vehicles and demand response. In E3S web of conferences(Vol. 87, p. 01007). EDP Sciences.
19. Zhang, Y., Gatsis, N., & Giannakis, G. B. (2013). Robust energy management for microgrids with high-penetration renewables. IEEE transactions on sustainable energy, 4(4), 944-953.
20. Reka, S. S., Dragicevic, T., Venugopal, P., Ravi, V., & Rajagopal, M. K. (2024). Big data analytics and artificial intelligence aspects for privacy and security concerns for demand response modelling in smart grid: A futuristic approach. Heliyon, 10(15).
21. Kaur, D., Islam, S. N., Mahmud, M. A., Haque, M. E., & Dong, Z. (2020). Energy forecasting in smart grid systems: A review of the state-of-the-art techniques. arXiv preprint arXiv:2011.12598.
22. Biswal, B., Deb, S., Datta, S., Ustun, T. S., & Cali, U. (2024). Review on smart grid load forecasting for smart energy management using machine learning and deep learning techniques. Energy Reports, 12, 3654-3670.
23. Rahman, R., Kumar, N., & Nguyen, D. C. (2025, January). Electrical load forecasting in smart grid: A personalized federated learning approach. In 2025 IEEE 22nd Consumer Communications & Networking Conference (CCNC)(pp. 1-2). IEEE.
24. Taïk, A., & Cherkaoui, S. (2020, June). Electrical load forecasting using edge computing and federated learning. In ICC 2020-2020 IEEE international conference on communications (ICC)(pp. 1-6). IEEE.
25. Ahamed, A., Tarafder, M. T. R., Rimon, S. T. H., Hasan, E., & Al Amin, M. (2024, October). Optimizing Load Forecasting in Smart Grids with AI-Driven Solutions. In 2024 IEEE International Conference on Data and Software Engineering (ICoDSE)(pp. 182-187). IEEE.
26. Alam, M. M., Hossain, M. J., Habib, M. A., Arafat, M. Y., & Hannan, M. A. (2025). Artificial intelligence integrated grid systems: Technologies, potential frameworks, challenges, and research directions. Renewable and Sustainable Energy Reviews, 211, 115251.
27. Ohanu, C. P., Rufai, S. A., & Oluchi, U. C. (2024). A comprehensive review of recent developments in smart grid through renewable energy resources integration. Heliyon, 10(3).
28. Aslam, S., Altaweel, A., & Nassif, A. B. (2023). Optimization algorithms in smart grids: A systematic literature review. arXiv preprint arXiv:2301.07512.
29. Badr, A. A., Saafan, M. M., Abdelsalam, M. M., & Haikal, A. Y. (2023). Novel variants of grasshopper optimization algorithm to solve numerical problems and demand side management in smart grids. Artificial intelligence review, 56(10), 10679-10732.
30. Melhem, F. Y. (2018). Optimization methods and energy management in" smart grids"(Doctoral dissertation, Université Bourgogne Franche-Comté).
31. Nakabi, T. A., & Toivanen, P. (2021). Deep reinforcement learning for energy management in a microgrid with flexible demand. Sustainable energy, grids and networks, 25, 100413.
32. Zahraoui, Y., Korõtko, T., Rosin, A., Mekhilef, S., Seyedmahmoudian, M., Stojcevski, A., & Alhamrouni, I. (2024). AI applications to enhance resilience in power systems and microgrids—A review. Sustainability, 16(12), 4959.
33. Shibi, B. (2025). AI-Driven Smart Energy Management Systems for Optimized Renewable Energy Utilization in Urban Smart Grids.
34. Ozay, M., Esnaola, I., Vural, F. T. Y., Kulkarni, S. R., & Poor, H. V. (2015). Machine learning methods for attack detection in the smart grid. IEEE transactions on neural networks and learning systems, 27(8), 1773-1786.
35. Gómez, Á. L. P., Maimó, L. F., Celdrán, A. H., Clemente, F. J. G., Sarmiento, C. C., Masa, C. J. D. C., & Nistal, R. M. (2019). On the generation of anomaly detection datasets in industrial control systems. IEEE Access, 7, 177460-177473.
36. He, H., & Yan, J. (2016). Cyber‐physical attacks and defences in the smart grid: a survey. IET Cyber‐Physical Systems: Theory & Applications, 1(1), 13-27.
37. Wu, J., Ota, K., Dong, M., Li, J., & Wang, H. (2016). Big data analysis-based security situational awareness for smart grid. IEEE Transactions on Big Data, 4(3), 408-417.
38. Galván, L., Navarro, J. M., Galván, E., Carrasco, J. M., & Alcántara, A. (2019). Optimal scheduling of energy storage using a new priority-based smart grid control method. Energies, 12(4), 579.
39. Kamankesh, H., Agelidis, V. G., & Kavousi-Fard, A. (2016). Optimal scheduling of renewable micro-grids considering plug-in hybrid electric vehicle charging demand. Energy, 100, 285-297.
40. Siano, P. (2014). Demand response and smart grids—A survey. Renewable and sustainable energy reviews, 30, 461-478.
41. Kermansaravi, A., Refaat, S. S., Trabelsi, M., & Vahedi, H. (2025). AI-based energy management strategies for electric vehicles: Challenges and future directions. Energy Reports, 13, 5535-5550.
42. Rodriguez-Calvo, A., Cossent, R., & Frías, P. (2018). Scalability and replicability analysis of large-scale smart grid implementations: Approaches and proposals in Europe. Renewable and Sustainable Energy Reviews, 93, 1-15.
43. Siddiqui, F., Zeadally, S., Alcaraz, C., & Galvao, S. (2012, July). Smart grid privacy: Issues and solutions. In 2012 21st International Conference on Computer Communications and Networks (ICCCN)(pp. 1-5). IEEE.
44. Onu, P., Pradhan, A., & Madonsela, N. S. (2024, June). Explainable AI (XAI) in Smart Grids for Predictive Maintenance: A survey. In 2024 1st International Conference on Smart Energy Systems and Artificial Intelligence (SESAI)(pp. 1-6). IEEE.