Improving LVRT Capability in Grid Connected PV System Using DRL-Based Controller

As the grid integration of renewable energy sources increases, improving the Low Voltage Ride-Through (LVRT) capability of these systems has become critical for grid stability and energy security. This paper focuses on achieving LVRT capability in a two-stage three-phase grid-connected photovoltaic (PV) system. Deep Reinforcement Learning (DRL) techniques address this challenge through their ability to adapt to dynamic and uncertain grid conditions. DRL stands out from traditional methods due to its capacity to generalize across varying fault scenarios while minimizing system errors and ensuring grid code (GC) compliance. A grid-connected PV system is modeled in MATLAB/Simulink and controlled using two DRL algorithms: Deep Deterministic Policy Gradient (DDPG) and Twin Delayed Deep Deterministic Policy Gradient (TD3). The performances of these DRL methods are evaluated against a conventional Proportional-Integral (PI) controller optimized with the Symbiotic Organisms Search (SOS) algorithm under three-phase balanced faults. Key performance metrics include t-test results, GC compliance, AC current and DC voltage limits, and error values under three-phase balanced faults conditions. The analysis shows that TD3 achieved 4.31% lower error than DDPG and 12.85% lower error than SOS-PI. The t-test analysis revealed a significant difference in ITAEav performance between TD3 and SOS-PI (p =0.0089), indicating TD3’s superior performance. DRL-based controllers effectively ensure GC compliance across all tested faults, while the SOS-PI controller struggles to generalize to different fault conditions. This highlights the necessity of DRL techniques for achieving reliable and robust LVRT capability in grid-connected PV systems.