Discrete-time super-twisting controller using neural networks

Abstract

This paper presents a neural super-twisting controller for a class of discrete-time nonlinear systems. A general plot of the control problem consists in considering a discrete-time nonlinear system with a block controllable structure, and assuming that the equations of the nonlinear system have virtual controls, a discrete-time super-twisting algorithm is applied such that each equation follows a desired trajectory until the real control is achieved. On the other hand, taking into account the capacity of neural networks to learn the behaviour of complex systems, it is proposed a neural network with a block controllable structure for identifying and controlling a discrete-time nonlinear system. Neural network weights are trained with the cubature Kalman filter algorithm. To show the effectiveness of the proposed neural super-twisting controller, a numerical simulation is performed on a Quanser 2-DOF helicopter.