Grid Import Optimization with Adaptive Deep Reinforcement Learning for PV-Battery Systems

This article explores the application of Deep Reinforcement Learning (Deep RL) to optimize energy management in photovoltaic (PV) and battery systems. The new framework presented here includes important innovations such as Rule-Based Action Smoothing for system performance consistency, PPO Multi-House Training to generalize across a wide range of energy usage patterns, and Post-Controller Integration to deal with real-time operational issues. While the dataset originates from Ireland, the model is adapted to align with Indonesia's dual-tariff system and local energy regulations. Simulation results demonstrate substantial cost savings, with reductions of up to 85.28% in stable scenarios and 18.26% in high-variability environments. These results highlight the flexibility and resilience of the methodology for using renewable energy to reduce costs and increase system efficiency. The model is, therefore, scalable for the implementation of intelligent energy systems in the residential context to support Indonesia's renewable energy goals and demonstrate its applicability to a broad range of scenarios.