A Study on Home Integrated Energy System Optimization Based on Deep Reinforcement Learning

This paper investigates the challenges posed by the high penetration of intermittent the integration of renewable energy with energy storage systems into traditional household energy management systems (HEMS). To cope with these challenges, we design a control framework founded on deep reinforcement learning (DRL) that optimizes the operation of batteries and heat pumps, with the dual objectives of lowering cost reduction and reducing renewable energy curtailment. A multi-objective reward function was designed, informed by expert knowledge, and an interactive simulation environment was developed for algorithm validation. The experimental evaluation reveals the superior performance of the TD3 algorithm over the benchmark approaches, resulting in a 12.61% reduction in operating expenses while significantly improving photovoltaic self-sufficiency. These findings highlight the potential of DRL in advancing intelligent energy management and provide guidance for algorithm selection across diverse application scenarios.