RL and Rule-Based Pricing Compared for P2P Energy Trading

This research investigates and compares the effectiveness of two distinct approaches for pricing electricity in peer-to-peer (P2P) trading within residential communities equipped with photovoltaic (PV) systems. The study evaluates both traditional rule-based pricing mechanisms and more advanced learning-based methods, specifically reinforcement learning (RL) implemented via a Deep Q-Network. Rule-based benchmarks included bill-sharing, mid-market rates, and supply-demand-ratio (SDR) pricing. The findings reveal that in scenarios where only PV generation is considered, the rule-based benchmarks generally delivered better community savings than the best RL policy. However, the introduction of battery energy storage (BES) dramatically altered the landscape. When BES was integrated, the community savings under the best RL policy saw a substantial increase, improving from EUR 734.23 to EUR 978.52. Across all learning-based configurations, SDR-shaped pricing consistently outperformed multiplier-based parameterizations. The paper concludes that while rule-based pricing remains highly competitive in simpler setups, the addition of energy storage significantly enhances the outcomes for learning-based approaches. Despite overall community benefits, the distribution of these financial advantages remained heterogeneous among individual households.