用神经网络进行连续时间非线性系统建模的研究

Neural Network Identification for Continuous Time Nonlinear Systems

在用神经网络进行系统建模时 ,建模误差的存在是难免的 .为了减小这种误差 ,本文对连续时间非线性系统提出了一种新的神经网络辨识模型 ,它是由带有输入修正的神经网络和稳定滤波器组合而成 .文中给出了权值的学习算法 ,即权值是根据辨识误差的投影算法来改变 。

In this paper a new identification model constructed by neural networks with modified inputs and stable filters is presented for continuous time nonlinear systems in order to reduce the inherent network approximation errors.An adaptive law with projection algorithm is employed to adjust the parameters of networks.Under certain conditions,convergence of the identification error is proved.