Combination of Artificial Neural Network-based Approaches to Control a Grid-connected Photovoltaic Source under Partial Shading Condition

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