Predictive LSTM–Reinforcement Learning Framework for Adaptive Energy Distribution in Solar–Wind Hybrid EV Systems

Abstract

Effective energy management in solar-wind hybrid electric vehicle (EV) systems is complicated by variable renewable supply, unpredictable EV demand, and changing grid pricing. Conventional forecasting and allocation techniques frequently struggle with handling real-time fluctuations, leading to excessive energy usage. This study presents a hybrid framework combining Long Short-Term Memory (LSTM) and Reinforcement Learning (RL) that integrates accurate short term energy generation forecasts with adaptive decision-making for optimal energy management. The LSTM module forecasts solar and wind generation utilising multivariate time-series data, encompassing meteorological and system characteristics, while the RL agent allocates energy dynamically among electric vehicles, batteries, and the grid. Simulation findings exhibit enhanced performance compared to baseline approaches, attaining 98.3% accuracy, 97.9% precision, 98.1% recall, 98.0% F1-score, a root mean square error (RMSE) of 1.9, and a R² of 0.99. Comparative analyses utilising Random Forest, independent LSTM, Deep Q-Network, and Support Vector Regression validate that the proposed framework enhances prediction accuracy, energy efficiency, and durability under variable settings. This study presents a scalable, real-time, and dependable solution for renewable energy management in electric vehicles, surpassing current methodologies and delivering actionable information for the sustainable implementation of smart grids.

https://doi.org/10.65470/james.5