Adaptive Recurrent Neural Network Enhanced Variable Structure Control for Nonlinear Discrete Mimo Systems

Abstract

Abstract The stability and performance of robust control for a nonlinear complex dynamic system deteriorates due to large uncertainties. To address this, a discrete variable structure control (DVSC) outside of a convex set is designed to force the operating point into a convergent set, which is verified by Lyapunov stability theory. Due to the dynamic features of the lumped uncertainties, they are learned on‐line by a recurrent neural network (RNN) in a specific convex set containing the convergent set of the DVSC. Between this convergent set and the convex set, a switching mechanism determines whether a learning RNN for the lumped uncertainties is executed. An adaptive recurrent‐neural‐network enhanced discrete variable structure control is then established to improve both transient and steady performances. Simulations, including compared examples and application to the trajectory tracking of a mobile robot, validate the effectiveness and robustness of the proposed control.