Analyses of Deep Reinforcement Learning and Conventional MPPT Control under Fast-Changing Irradiance

Abstract

This paper investigates a deep reinforcement learning (DRL) maximum power point tracking (MPPT) strategy for a photovoltaic (PV) boost converter system using the Deep Deterministic Policy Gradient (DDPG) algorithm. The proposed controller is designed to overcome limitations of conventional perturb-and-observe (P&O) methods combined with PID control, particularly under non-uniform irradiance and load impedance variations. The DDPG agent is trained offline and learns the nonlinear mapping between PV electrical states and duty-cycle control, while explicitly accounting for DC-link voltage regulation. A comparative performance evaluation is conducted against conventional P&O–PID and a PSO–NN MPPT scheme under fast irradiance transients, and varying load conditions. Simulation results show that while the P&O–PID and PSO–NN controller achieves marginally higher instantaneous PV power under rapid irradiance changes, the proposed DRL controller provides superior DC-bus voltage regulation and sustained stability within a simulated irradiance levels ranging from 1000 W/m2 to 400 W/m2. This reflects a trade-off between energy extraction and system-level stability but overall, the DRL MPPT approach demonstrates improved robustness under fast environmental transients and realistic operating conditions, highlighting its suitability for standalone DC microgrid and advanced PV power conversion applications.