Improved Performance for the DC-AC Converters Control System Based on PCH Controller and Reinforcement Learning Agent

Starting from the classical structure of a three-phase voltage DC-AC converter whose basic controller is designed based on the PI-type control law, this article shows the structure of a DC-AC converter control system (CCS) based on the Port Controlled Hamiltonian (PCH) controller, along with the improvement of DC-AC CCS performance by means of machine learning (ML) strategy. Among these strategies, the most suitable for process control is reinforcement learning (RL), and the RL Twin-Delayed Deep Deterministic Policy Gradient (TD3) agent was chosen from the concrete implementations. The control structures and the synthesis of the PCH control law based on passivity theory are presented, and, in addition, the creation and training of an RL-TD3 agent is presented. Through numerical simulations it is proved the improvement in the DC-AC CCS performance in case of using the RL-TD3 agent in terms of the performance indicators of the control systems, of which we mention: response time, steady-state error, ripple, but also in terms of the quality of electricity according to the Total Harmonic Distortion (THD) analysis.