A Computationally Efficient and Stable Learning-Based Controller for DC/AC Inverter

Abstract

A new machine-learning-based control strategy for the dc/ac inverter is proposed in this article, which is highly computationally efficient and insensitive to model parameter variations. It provides fast offline neural network (NN) training and low computational cost for online digital signal processor (DSP)-based implementation. The recurrent equilibrium network (REN) is employed to achieve excellent transient and steady-state performance and the closed-loop system is proven to be asymptotically stable. The performance of the proposed approach is verified through experimental comparisons with the existing control methods.