Predefined-time adaptive neural network decentralized control for large-scale interconnected systems with input hysteresis

Abstract

This study endeavors to develop a predefined-time adaptive neural network decentralized controller for large-scale interconnected nonlinear systems with input hysteresis. Within the framework of the backstepping technique, the proposed control scheme guarantees that the tracking error converges to a small bounded set within a predefined settling time. The upper limit of this convergence time is determined by a single adjustable control parameter. Modified command filter not only tackles the inherent “complexity explosion” issue in traditional backstepping methods but also effectively avoids chattering phenomena possibly induced by $sign$ function. An online approximator based on neural networks is utilized to address system uncertainties. Moreover, a novel predefined-time error compensation mechanism is constructed to compensate for the reduction in control accuracy caused by filtering errors. Two simulation case studies demonstrate the feasibility and effectiveness of the proposed control method.