Maximum power point tracking of solar - pv under partial shading and variable load resistance using deep reinforcement learning algorithm

This study develops a deep reinforcement learning (DRL) maximum power point tracking (MPPT) strategy for photovoltaic (PV) systems operating under uniform irradiance and partial shading conditions. A Deep Deterministic Policy Gradient (DDPG) controller is designed to directly regulate the duty cycle of a PV boost converter, enabling continuous, model-free tracking of the global maximum power point (GMPP). Unlike conventional MPPT techniques that rely on predefined perturbation rules or averaged system models, the proposed approach formulates MPPT as a physically constrained control problem at the switching-converter level. A custom reward function is formulated to simultaneously maximize power extraction and penalize abrupt duty-cycle variations, thereby improving converter safety and operational robustness. The DDPG agent is trained using diverse curved and distorted irradiance profiles and variable load conditions to enhance policy generalization. Performance is benchmarked against classical P&O–PID control and advanced nonlinear intelligent MPPT strategies as well as real life case study. Under uniform irradiance, the proposed controller achieves smooth convergence with the lowest voltage root mean square error, while under partial shading it consistently tracks the GMPP, delivering up to 10% higher power output with ripple below 0.3% compared to conventional PID control. The DDPG–DRL controller demonstrates superior stability and robustness under a combination of distorted irradiance and high-load resistance scenarios, maintaining ripple below 0.2% and voltage operation closer to theoretical optimal points. These results confirm the effectiveness and practical applicability of the proposed DDPG–DRL MPPT framework for real-world photovoltaic energy systems.