Maximum Power Point Tracking Based Proximal Policy Optimization Algorithm for Grid-Connected Photovoltaic Systems

Photovoltaic (PV) systems are widely acknowledged for their sustainable, plentiful, and eco-friendly energy generation. However, the non-linear behavior of PV cells highlights the vital role of maximum power point tracking (MPPT) methods in optimizing their performance. Traditional MPPT techniques often struggle with partial shading conditions (PSCs). To improve MPPT effectiveness, incorporating artificial intelligence methods like deep reinforcement learning (DRL) can be advantageous. This study introduces a crayfish optimization algorithm (COA) based on DRL, specifically proximal policy optimization (PPO), designed to trail the maximum power point (MPP) of a 1-MW grid-connected PV system that includes an energy storage system (ESS) to alleviate fluctuations in PV output power. The control of the ESS is also optimized using another PPO controller. Additionally, the DRL controllers modify the measured PV voltage and ESS current according to reference values obtained from the COA method and variations in PV output. Simulation results using MATLAB/Simulink demonstrate the superior performance of the proposed MPPT and DRL controllers when compared to the traditional perturb and observe algorithm combined with a proportional-integral controller (P&O-PI). Moreover, the obtained results indicated that the COA-PPO method attained a DC boost converter efficiency of 95.14% and a PCC efficiency of 97.24%. In comparison, the PO-PI method achieved efficiencies of 94.24% and 96.30%, respectively.