Evolving LLM-Derived Control Policies for Residential EV Charging and Vehicle-to-Grid Energy Optimization

This research presents a novel application of Evolutionary Computation (EC) to residential electric vehicle (EV) energy management. While reinforcement learning (RL) achieves high performance in vehicle-to-grid (V2G) optimization, it produces opaque "black-box" neural networks difficult for consumers and regulators to audit. We propose a program search framework leveraging Large Language Models (LLMs) as intelligent mutation operators within an iterative prompt-evaluation-repair loop—analogous to a (1+1) Evolution Strategy—using the EV2Gym simulation environment as a fitness function. The system synthesizes executable Python policies balancing profit maximization, user comfort, and physical safety. We benchmark four prompting strategies—Imitation, Reasoning, Hybrid, and Runtime—evaluating their ability to discover adaptive control logic. The Hybrid strategy produces concise, human-readable heuristics achieving 118% of baseline profit, discovering complex behaviors like anticipatory arbitrage and hysteresis without explicit programming. This establishes LLM-driven EC as a practical approach for generating transparent, inspectable EV charging policies suitable for real residential deployment.