A Study on Residential PV-Battery System Optimization Based on DDPG Algorithm

Abstract

The integration of distributed photovoltaic (PV) systems and energy storage systems (ESS) creates challenges for residential energy management due to the stochastic nature of renewable energy sources (RES) and load demand. Traditional rule-based control strategies often struggle to capture the full economic benefits under dynamic pricing. To address this limitation, this paper proposes a continuous control framework utilizing the Deep Deterministic Policy Gradient (DDPG) algorithm to optimize the operation of grid-connected residential systems. By formulating a multi-objective reward function that balances electricity costs against PV self-consumption, the proposed agent learns to generate precise continuous actions for battery scheduling. Numerical simulations using real-world data from a smart community in Japan demonstrate that this approach reduces annual operational costs by 15.32% compared to a baseline strategy while simultaneously maintaining a high level of RES utilization. These findings highlight the potential of deep reinforcement learning (DRL) in advancing intelligent Home Energy Management System (HEMS) towards greater autonomy and efficiency.