A composite neural network-based adaptive sliding mode control method for reluctance actuator maglev system

To achieve high levitation control performances for a maglev system (MLS) with the uncertainties caused by the inherent nonlinearities and external disturbances, this paper proposes a novel composite adaptive sliding mode control (CASMC) method. The CASMC method comprises an equivalent controller, a composite neural network (NN) compensator, and a composite adaptive switching controller. Firstly, a modified prediction error that adds an adaptive switching term is designed to enlarge the effect of the compensation error. Secondly, a composite weight updating law consisting of the modified prediction error and the sliding mode surface is used for NN to accelerate its convergence speed. Thirdly, a new composite adaptive switching control law, including a prediction error-based adaptive switching gain and a prediction error-based proportion switching gain, is proposed for better dynamic response, stronger disturbance suppression capability, and lower chattering. The stability of the closed-loop control system is analyzed by the Lyapunov theorem. Comparative experiments were performed. Results show that the CASMC method can guarantee high levitation control performances with a better dynamic response, stronger robustness, no overshoot, and lower chattering simultaneously.