Beyond dynamic pricing: Discovering fixed, deep reinforcement learning-derived charging policies for electric vehicles

The increasing popularity of EVs threatens grid stability, increasing the costs for the electric distribution industry. While many studies propose complex dynamic pricing schemes that are difficult to implement and communicate to consumers, we introduce a fundamentally different paradigm: using Deep Reinforcement Learning (DRL) as a discovery tool—not for deployment, but to extract practical, fixed Time-of-Use (TOU) pricing policies. Unlike prior work that deploys DRL for real-time pricing, our approach leverages DRL’s learning capability to discover robust static policies. Our goal is to minimize intra-day load fluctuations while preserving user autonomy and avoiding the implementation challenges of daily-changing prices. We developed a detailed agent-based simulation to model realistic EV driver behavior across 12 distinct charging infrastructure scenarios, varying the availability of home, work, and public chargers. We demonstrate that our proposed fixed TOU policy — derived from the median prices of the dynamically trained DRL agent — is highly effective. This static policy significantly reduces load profile volatility, decreasing the average intra-day standard deviation by 18.03% compared to a constant pricing baseline and outperforming the most competitive of four dynamic heuristic benchmarks by 9.7%. This study confirms that a DRL framework is a powerful tool for discovering robust, effective, and easily implementable pricing strategies for real-world grid management.