A class designed of variable structure controller based on neural network

Abstract

Variable structure control is a control method which is proposed in the fifties of the last century. It is apply to the nonlinear, time delay, MIMO system. And it has the adaptive ability to the system disturbance and the system perturbation. Therefore, the variable structure control has been a research hot. But variable structure control has some defects, that is chattering phenomenon in the control process. How to reduce and eliminate the chattering is the main research direction on variable structure control. The neural network has learning ability, adaptive and can approximate any nonlinear function etc. The paper built the control object model through the on-line identification method on the base of neural network. Associate the predict value with the reference value, we can establish a performance index function. Then we can adjust the parameters of the variable structure controller to reduce the chattering phenomenon through the back propagation algorithm. This paper takes the self-balance robot as an experimental object, design a variable structure control bases on the neural network, and compare it to the traditional variable structure control. The simulation results show that the new control can reduce the chattering phenomenon than the traditional method effectively an, has better control accuracy and robustness.