Adjusting the Dynamic Frequency Response in Photovoltaic-Energy Storage Grid-Connected Systems with the Help of Deep Deterministic Policy Gradient

Abstract

This study applies the DDPG (Deep Deterministic Policy Gradient) algorithm to optimize the dynamic frequency response of PV (photovoltaic)-ESgrid-connected systems under high-dimensional continuous control. A high-fidelity model incorporating PV output, ES SOC (State of Charge), and frequency dynamics was developed from operational data. An actor-critic architecture with target networks and experience replay trains the control strategy for precise ES charging/discharging. Compared to PID (Proportional-Integral-Derivative), fuzzy control, DQN (Deep Q-Network), and PPO (Proximal Policy Optimization), DDPG reduces maximum frequency deviation to 0.45 Hz, lowers average oscillation amplitude to 0.52 Hz, delivers a 7.1 s response, and achieves 85% charging/discharging efficiency and 75% ES utilization, extending SOC safe-range duration. DDPG enhances system stability and offers a cost-effective solution for new-energy frequency regulation.