Feedback Recurrent Equilibrium Network-based Linear Quadratic Regulator for DC/AC Inverter

This paper proposes a machine learning-based controller for the DC/AC inverter, integrating a linear quadratic regulator (LQR) with recurrent equilibrium networks (RENs). Traditional LQR controller suffers from model dependency and the difficulty in determining the cost function weight matrix. To address these issues, the proposed controller introduces an auxiliary control variable via feedback REN, ensuring superior transient and steady-state performance under varying operating conditions, including step changes in output current references, filter parameter uncertainties, and different load scenarios. The proposed approach is characterized by high computational efficiency during both training and implementation, with verified closed-loop system stability. The effectiveness of the proposed approach is validated through experimental comparisons with existing control methods.