Combination of artificial neural network-based approaches to control a grid-connected photovoltaic source under partial shading condition

This paper proposes an approach based on artificial neural networks (ANN) to control a grid-connected photovoltaic system (PVS) under partial shading (PS) conditions. In PS conditions, the P-V curve exhibits multiple peaks, with only one representing global maximum power point (GMPP), and others local points (LMPP). Traditional Maximum Power Point Tracking (MPPT) methods are unable identify GMPP get stuck around LMPP, which results in reduced productivity of PVS. The proposed combines supervised learning (SL) deep reinforcement (DRL) techniques design controller hierarchical structure that can overcome problem identifying PVSs PVS study consists four identical solar panels. At first level, each panel has sub-controller designed using ANN SL technique, determines appropriate duty cycle extract from real-time weather second DRL agent identifies optimal for DC/DC converter cycles generated by sub-controllers. Deep Deterministic Policy Gradient (DDPG) Twin Delayed DDPG (TD3) agents implemented evaluated level control. Simulation MATLAB/Simulink demonstrate effectiveness tracking GMPP.