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

Abstract

Analyzing the problem of connecting a microgrid to the main grid by means of a voltage source inverter while maintaining a steady voltage in case with variation of load in the form of balanced/unbalanced linear resistance or nonlinear resistance, this paper presents, based on robust systems theory, the synthesis of a robust controller that achieves the above-mentioned goal. A maj or role is played by the network filters, the weights associated with the extended robust system together with the chosen topology. A combined control system based on a robust controller and a Reinforcement Learning-Twin-Delayed-Deep-Deterministic-Policy-Gradient (RL- TD3) agent is presented to improve the control system performance of this DC-AC converter. The RL- TD3-type agent is trained, tested and validated, and after implementation in the control system it is able to provide correction signals for the robust control in order to achieve superior performance in terms of steady-state error, ripple and Total Harmonic Distortion (THD). The numerical simulations confirm the superiority of the DC-AC converter control system using the robust controller combined with RL- TD3 agent.