Research On System Identification Method Based On Neural Network

Many functions are possessed by the neural network that is, parallel processing, self-learning and self-adapting. It could approximate any nonlinear function with any precision. Therefore, the neural network is widely used in many aspects: pattern recognition, system identification and control fields and so on. It is developed in the paper about the application of neural network in pattern recognition and system identification, several novel methods are given for some systems identification and the parameters identification.Considering a class of stochastic system under noise disturbance, by the discretion of the system error bounds and transforming the system identification problems to pattern recognition problems, this paper presents a description method to system model, and the corresponding neural network model has been built. The model makes full use of the known probability information of disturbance and can efficiently simulate the probability distribution information of system outputs. The identification results are more visible and usable. Besides, the identification model can be quickly established to provide a feasible way for on-line identification of stochastic system.With the concept of complete state point space, a new method of neural networks ensemble is proposed. By guarantying the accuracy and generalization ability of neural networks ensemble identification, the method combines both the type and parameters of the systems and lowers the commands to information quantity of the object.