Reinforcement Learning-Based Charging Control in Households Under Real-Time Pricing

Abstract

This paper presents a reinforcement learning-based approach for real-time charging control in households under realtime pricing. A Soft Actor-Critic agent is trained to optimize the charging and discharging of a stationary battery to minimize electricity costs while ensuring operational stability. Unlike predictive or rule-based approaches, the RL agent makes decisions without access to future prices, relying solely on current market conditions and past observations. The methodology is tested using historical market data from three electricity markets with distinct price dynamics: Germany, Austria, and Italy. Results demonstrate that the RL-based control successfully adapts to different market structures, effectively leveraging price fluctuations to improve household energy management.