用神经网辨识非线性系统中的模型误差分析(I)——函数逼近误差及其对确定性动态系统的影响

Model Error Analysis in Nonlinear System Identification Using Neural Networks(I)-Function Approximation Error andIts Effect on Deterministic Dynamic System

讨论Sigmoid型三层前馈网的最佳逼近特性和插值特性,给出了网络实现精确插值的条件;分析了网络逼近误差及运行初始误差对网络模型的影响,指出在一定条件下网络模型输出与实际系统输出相差不大。

The effect of noise on the model multi-step prediction is discussed for identifying a stochastic sys-tem including simple system noise- The drawback of the series - parallel scheme used in system including simple output noise is pointed out and the arbitrary approximation property under the parallel scheme is proved.