本文针对用于复杂非线性辨识的连续激发函数的神经网络,提出一种根据被测对象非线性特性设计网络初值的算法,算例及分析表明,这种新赋初值算法不仅能使网络的收敛速度有一定提高,辨识误差有所下降,且可避免收敛于局部极小点。 Aiming at the neural network of continuous excitation function for complex nonlinear identification, this paper proposes an algorithm to design the network initial value according to the nonlinear characteristics of the measured object. The examples and analysis show that this new initial value algorithm can not only make the network The convergence rate has a certain increase, identification error decreased, and can avoid convergence in the local minimum.