Explainable Machine Learning for Predicting Electric-Vehicle (EV) Charging Demand in Renewable-Energy-Powered Grids: A Comprehensive Review

Abstract

The rapid global adoption of electric vehicles (EVs) coupled with the increasing integration of renewable energy, presents new challenges in managing electricity demand, ensuring grid stability and achieving decarbonization targets. Accurate forecasting of EV charging demand has become vital for optimizing infrastructure deployment, dynamic pricing and renewable integration. This paper presents a comprehensive review of explainable machine leaning (ML) approaches for EV charging demand forecasting under renewable energy powered grids. The review categorizes methods into traditional statistical models ensemble-based techniques deep learning architectures and hybrid approaches. Emphasis is placed on explainable artificial intelligence (XAI) techniques particularly Shapley Additive Explanations (SHAP) which enhance interpretability by identifying the key features driving model predictions. Results from recent studies indicate that renewable energy penetration, electricity price and carbon intensity are the dominant factors influencing charging demand. The review highlights how explainable ML models can support policymakers in designing demand-response strategies grid operators in balancing supply and demand and consumers in making environmentally conscious charging decisions. Finally research challenges and future opportunities are discussed, including the need for causal interpretability, real-time adaptation and data privacy in federated learning frameworks.

Introduction

The electrification of transportation is central to global decarbonization efforts, with the International Energy Agency (IEA) projecting over 350 million electric vehicles (EVs) worldwide by 2030. While EV adoption significantly reduces greenhouse gas emissions, it introduces operational challenges for power systems due to the spatio-temporal variability of charging demand and the intermittency of renewable energy sources such as solar and wind. Traditional grid models, built on predictable load assumptions, struggle to handle decentralized renewable integration, leading to potential supply–demand imbalances, voltage instability, and increased operational costs.

Accurate EV charging demand forecasting is therefore critical for maintaining grid reliability, optimizing tariffs, allocating resources efficiently, and supporting renewable integration.

Machine Learning for EV Charging Forecasting

Forecasting approaches have evolved from traditional statistical models such as ARIMA and multiple linear regression to advanced machine learning (ML) and deep learning techniques.

Key Model Categories:

1. Linear/Statistical Models (ARIMA, MLR)

   • Simple and interpretable

   • Limited ability to model nonlinear and dynamic relationships

2. Tree-Based Models (RF, XGBoost, LightGBM, CatBoost)

   • High prediction accuracy

   • Handle multidimensional features effectively

   • Interpretable when combined with explainability tools

3. Deep Learning Models (LSTM, GRU, CNN, Seq2Seq)

   • Capture temporal and spatial dependencies

   • Suitable for multi-step and real-time forecasting

   • Require large datasets and have explainability challenges

4. Graph Neural Networks (GCN, ST-GNN)

   • Model spatial dependencies among charging stations

   • Reduce forecasting errors in urban networks

   • Computationally intensive and less interpretable

5. Hybrid & Reinforcement Learning Models

   • Integrate prediction with optimization and demand-response control

   • Support smart-grid management

   • More complex to design and interpret

Among these, ensemble models like XGBoost and LightGBM dominate practical applications due to their balance of performance, scalability, and interpretability.

Renewable Energy Integration

As renewable penetration increases, forecasting models must account for solar irradiance, wind speed, temperature, carbon intensity, and renewable share.

Studies show:

• Higher renewable penetration increases EV charging during daylight hours due to lower electricity prices.

• Charging demand shifts in response to dynamic pricing and carbon-aware policies.

• Aligning EV charging with renewable peaks reduces grid strain and improves sustainability.

Hybrid frameworks combining ML with optimization techniques and reinforcement learning enable adaptive demand-response strategies for smart-grid systems.

Explainable AI (XAI) in EV Forecasting

While ML models offer strong predictive performance, interpretability is essential for policy-making and grid management. Explainable Artificial Intelligence (XAI) tools improve transparency and stakeholder trust.

SHAP (Shapley Additive Explanations)

SHAP has become a standard interpretability method. It:

• Quantifies each feature’s contribution to a prediction

• Ensures fair attribution using cooperative game theory

• Satisfies properties of local accuracy, missingness, and consistency

SHAP analysis in EV forecasting reveals that:

• Renewable energy share

• Carbon intensity

• Electricity pricing

• Charging efficiency

are dominant factors influencing EV charging behavior.

Notably:

• Increased renewable penetration boosts midday charging.

• Higher carbon intensity discourages charging.

• Interaction between renewable share and pricing amplifies demand effects.

TreeSHAP enables efficient computation for tree-based models like XGBoost, enhancing interpretability without sacrificing speed.

Comparative Insights

• Ensemble methods provide strong balance between accuracy and interpretability.

• Deep learning models outperform others when large temporal datasets are available.

• Graph neural networks excel in complex urban networks with spatial interdependencies.

• However, deep and graph-based models require XAI integration to ensure transparency.

Hybrid models combining ML with reinforcement learning represent a growing trend, offering multi-objective optimization that balances accuracy, efficiency, and sustainability.

Renewable Energy & EV Charging Symbiosis

EV charging and renewable integration are interdependent:

• Aligning EV charging with solar peaks can reduce grid load by up to 25%.

• Night-time wind-aligned charging improves energy utilization by 15%.

• Carbon-aware tariff design can shift charging to cleaner generation periods.

This synergy supports sustainable grid operation and net-zero energy goals.

Conclusion

This comprehensive review has synthesized developments in machine-learning-based forecasting of EV charging demand with particular attention to the role of explainable artificial intelligence (XAI) in achieving transparent, sustainable and policy-relevant outcomes. The survey of over forty peer-reviewed studies demonstrates that: • Tree-based ensembles (XGBoost, LightGBM, CatBoost) remain practical and interpretable for short-term forecasting. • Deep sequence models (LSTM, GRU Transformers) outperform for long-horizon predictions but require interpretability augmentation via SHAP or Integrated Gradients. • Graph neural networks (GNNs) have achieved state-of-the-art performance for urban station-level forecasting by capturing spatial and topological dependencies. • Explainability frameworks, especially SHAP have proven indispensable in revealing the dominant impact of renewable share, carbon intensity and electricity pricing on charging behavior. • Hybrid and probabilistic models offer promise for risk aware real time energy management. Future research should prioritize causal interpretability, uncertainty quantification, federated learning and ethical transparency. As EV adoption accelerates and renewable integration deepens explainable ML forecasting will become a cornerstone for next generation smart grids and carbon-neutral mobility ecosystems.

References

[1] T. Zhang, Q. Peng, S. Zeng, “Predicting EV Charging Demand in Renewable-Energy-Powered Grids Using Explainable Machine Learning,” Sustainability, vol. 17, no. 9, 4158, 2025. doi: 10.3390/su17094158 [2] Y. Li et al., “Impact of Renewable Integration on Load Forecasting,” IEEE Trans. Sustain. Energy, 2024. [3] S. Wang et al., “Predicting Electric Vehicle Charging Demand Using Graph Neural Networks,” Transp. Res. Part C, 2023. [4] M. Alaraj et al., “Machine Learning-Based Electric Vehicle Charging Demand Forecasting,” Energies, 2025. [5] A. Mystakidis et al., “EV Charging Forecasting Exploiting Traffic, Weather and User Behavior,” Int. J. Mach. Learn. Cybern., 2025. [6] S. Lundberg and S.-I. Lee, “A Unified Approach to Interpreting Model Predictions,” Proc. NeurIPS, 2017. [7] Y. Li et al., “Federated Graph Learning for EV Charging Demand Forecasting,” arXiv:2405.00742, 2024. [8] A. K. Jain et al., “Short-Term Load Forecasting: A Review,” Electr. Power Syst. Res., 2022. [9] S. Fan and L. Chen, “Probabilistic Electric Load Forecasting: A Tutorial,” IEEE PES Mag., 2019. [10] T. Chen and C. Guestrin, “XGBoost: A Scalable Tree Boosting System,” Proc. KDD, 2016. [11] G. Ke et al., “LightGBM: A Highly Efficient Gradient Boosting Decision Tree,” NeurIPS, 2017. [12] A. Prokhorenkova et al., “CatBoost: Unbiased Boosting with Categorical Features,” NeurIPS, 2018. [13] Y. Zhang et al., “Traffic-Aware EV Charging Prediction Using LSTM,” Transp. Res. Rec., 2023. [14] R. Gunasekaran et al., “Prediction of EV Charging Demand Using Deep Learning,” SN Appl. Sci., 2024. [15] A. Roozkhosh et al., “Predicting Electric Vehicle Demand for Charging Stations,” CaGIS/UCGIS, 2024. [16] H. Li et al., “UrbanEV: An Open Benchmark Dataset for Urban EV Charging,” Patterns, 2025. [17] P. Chen et al., “EV Charging Demand Forecasting at Charging Stations: Climatic Factors and Prediction Models,” IET Power Electron., 2025. [18] A. Neubauer et al., “Explainable Multi-Step Heating Load Forecasting,” Patterns, 2025. [19] L. Baur et al., “Explainability and Interpretability in Electric Load Forecasting,” Patterns, 2024. [20] M. A. Yassen et al., “Explainable AI for Wind Power Forecasting,” Neural Comput. Appl., 2025. [21] E. Gomes et al., “Explainable AI in Solar PV Mapping with SHAP,” Energies, 2025. [22] A. Ostrometzky et al., “Probabilistic Forecast of EV Charging Demand,” Energy Informatics, 2024. [23] M. Makridakis et al., “M5 Accuracy Competition: Forecasting with Machine Learning,” Int. J. Forecast., 2022. [24] A. Hong et al., “Carbon-Aware Electricity Markets and Dynamic Pricing,” Energy Policy, 2023. [25] D. Deka et al., “Learning for Distribution System Operations,” Proc. IEEE, 2023. [26] S. Wang, “Charging Demand with Trajectory Data,” Transp. Res. Part C, 2023. [27] IRENA, “Global Renewables Outlook,” 2024. [28] California Energy Commission, “Zero-Emission Vehicle Infrastructure Plan,” 2024. [29] A. Safdarian et al., “Electric Vehicle Load Modeling and Forecasting,” IEEE Trans. Smart Grid, 2020. [30] Z. Yang et al., “Hybrid CNN-BiLSTM for Very Short-Term Power Load Forecasting with SHAP,” Eng. Appl. Artif. Intell., 2025. [31] T. van Etten et al., “Large-Scale Forecasting of EV Charging,” Proc. INSTICC ICPRAM, 2024. [32] Y. Mao et al., “EV Charging Demand Forecasting: Data-Driven Integrated Learning on High-Dimensional Station Data,” Renew. Energy, 2025. [33] A. Roozkhosh et al., “Spatio-Temporal Forecasting for EV Charging Demand,” IEEE Access, 2025. [34] M. Alaraj et al., “Optimizing EV Charging Schedules with Renewable Energy Availability,” Energies, 2025. [35] Y. Chen et al., “EV Charging Demand Prediction Based on Spatio-Temporal Attention GCN,” Energies, 2025. [36] A. Mystakidis et al., “Integrating V2G Systems for Renewable Balancing,” Int. J. Energy Res., 2025. [37] M. Makridakis et al., “Benchmarking Energy Forecasting Competitions,” Int. J. Forecast., 2023. [38] H. Li et al., “UrbanEV: An Open Benchmark Dataset for EV Charging,” Patterns, 2025. [39] P. Chen et al., “EV Charging Behavior Modeling Across Regions,” IEEE Trans. Smart Grid, 2024. [40] S. Wang et al., “Adaptive Spatio-Temporal Graph Recurrent Network for EV Charging Forecasting,” Appl. Energy, 2025

Copyright

Copyright © 2026 Mariam Daood, Ajaz Husain Warsi, Shikha , Dr. Kaneez Zainab. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.