Deep Reinforcement Learning Based Control of a Grid Connected Inverter With LCL-Filter for Renewable Solar Applications

Abstract

This research paper presents a novel approach to current control in Grid-Connected Inverters (GCI) using Deep Reinforcement Learning (DRL) based Twin Delayed Deep Deterministic Policy Gradient (TD3) method. The study focuses on addressing the limitations of traditional control techniques and state of the art techniques, particularly Proportional-Integral (PI) control and Model Predictive Control (MPC), by leveraging the adaptive and autonomous learning capabilities of DRL. The proposed novel modified TD3-based DRL method learns an optimal control policy directly from raw data, enabling the controller to adapt and improve its performance in real-time. The research includes a comprehensive analysis of the implementation and validation of the modified TD3-based DRL control in a grid-connected three phase three level Neutral Point Clamped (NPC) inverter system with Inductor-Capacitor-Inductor (LCL) filter. Real-time validation experiments are conducted to evaluate the control performance, power transfer capability in grid compliance. Furthermore, a detailed comparison is presented with experimentation, highlighting the advantages of the TD3-based DRL control over PI and MPC control techniques. Robustness checking is performed under various operating conditions, including parameter variations and dynamic conditions in the grid. The results analysis demonstrates that the TD3-based DRL control outperforms traditional PI control techniques in terms of static, dynamic response, and robustness. Additionally, The DRL based grid connected inverter current control method is validated in Renewable Energy Source (RES) solar PV grid integration application.

Keywords

• Current control technique
• deep reinforcement learning
• grid-connected inverters
• model predictive control
• PI control
• real-time validation