Adaptive Control, Nonlinear Multi-model Based On Neural Networks

In this dissertation, a kind of nonlinear multiple models adaptive controller based on artificial neural network is designed for a class of SISO nonlinear discrete time system and a class of MIMO nonlinear discrete time. The stability of the control system is proven too. At first the nonlinear system is expanded to get a fixed linear model and a fixed nonlinear model at equilibrium point by Taylor expansion. Then a free running linear adaptive model and a free running nonlinear adaptive model based on neural network are designed, following with a reinitialized linear adaptive model and a reinitialized nonlinear neural network adaptive model. At each time, the optimal model and corresponding controller are selected according to the switching index. The stability of the system is proven.Simulation results illustrate that when operating conditions changes frequently, the method proposed in this dissertation is able to improve the system transient performance than the normal adaptive method.