Dynamic renewable energy integration for EV charging via model-based reinforcement learning

The rapid emergence of Electric Vehicles (EVs) presents both opportunities and challenges for sustainable transportation, particularly in integrating renewable energy sources into charging infrastructure while maintaining grid stability. Current EV charging systems exhibit inefficient energy utilization, peak demand stress, and suboptimal integration of intermittent renewable energy sources, primarily due to their reliance on conventional grid power and the lack of intelligent distribution mechanisms. Hence, the research introduces Smart Renewable Optimization with Vehicle-aware Reinforcement Learning (SRO-VRL), a model-based RL approach that makes real-time, adaptive decisions on energy distribution. This model dynamically optimizes EV charging schedules by prioritizing renewable energy sources, EV demand, and grid constraints to ensure efficient charging. The model begins by collecting real-time data from EV charging stations, grid load measurements, and renewable energy sources. The approach relies on collecting real-time data from private and public EV charging stations connected to local distribution grids at the community/microgrid level in Europe. This infrastructure includes mixed residential chargers (Type-2/Type-3 AC) and community charging stations. Its hybrid renewable energy mix, which is supported by battery energy storage, primarily comes from solar energy. The local utility system supplies additional power. Grid data show low-voltage feeders that are constrained by capacity and pricing. In comparison to benchmark reinforcement learning and rule-based strategies, the results demonstrate the effectiveness and scalability of the proposed SRO-VRL framework for community-scale microgrids and regional smart grids, improving renewable energy utilization by approximately 4–5%, lowering peak grid load by approximately 7–8%, and reducing overall EV charging costs by approximately 5–6%. System states are then formulated to encompass EV state-of-charge (SoC), predicted renewable supply, and grid constraints, while charging power allocations constitute the action space. In comparison to benchmark reinforcement learning and rule-based strategies, the results demonstrate the effectiveness and scalability of the proposed SRO-VRL framework for community-scale microgrids and regional smart grids, improving renewable energy utilization by approximately 4–5%, reducing peak grid load by approximately 7–8%, and lowering overall EV charging costs by approximately 5–6%.