A novel adaptive neural sliding mode control for systems with unknown dynamics

In this paper, an adaptive neural sliding mode controller (ANSMC) is proposed as an asymptotically stable robust controller for a class of Control Affine Nonlinear Systems (CANSs) with unknown dynamics. In the proposed method a Control Affine Radial Basis function Network (CARBFN) is developed for online identification of CANSs. A recursive algorithm based on Extended Kalman Filter (EKF) is used for training of CARBFN to develop an adaptive model for CANSs with unknown and uncertain system dynamics to reduce the uncertainties to low values. Since the CARBFN model learns the system time-varying dynamics online, the ANSMC will compute an efficient control input adaptively. Due to high degree of robustness, the proposed controller can be widely used in real world applications. To demonstrate this efficiency, a robust control system is successfully designed for a chaotic Duffing forced oscillator system in the presence of unknown dynamics as well as the unknown oscillation disturbance which is not available for measurement.