RBFNN Predictive Control Theory And Its Application In Satellite Attitude Simulation System

With the development of the aerospace technology, there is an increasing demand of the reliability and stable state precision. Aiming at the special features of the object in this field such as complicated structure and serious nonlinear, a neural networks predictive control strategy is designed and is applied to the modeling and control of the satellite attitude simulation system. RBF (Radial Basis Function) NN is more suitable for system identification and control because its outputs are linear to the weights. In this thesis, a novel online training algorithm for RBF NN is presented. The control laws and the update laws of NN weights are based on Lyapunov function. So the stability of the closed loop systems and the boundness of weights and errors are guaranteed. First, a dynamic neural network is constructed to simulate the dynamic behavior of the process by containing the past inputs and outputs into the input layer of the RBF network. And an online adaptive identification method is presented, which is useful for improving the reliability and simplifying the structure by using the given knowledge. Based on this dynamic model an internal model controller is designed, and the implemental method of object-positive model-reversal system is introduced, which is proved efficient.At last, the proposed scheme is carried out on the satellite attitude simulation system. The experiments show that the method has better performance and powerful robustness, which will direct the future application.