神经网络并联辨识算法的收敛性研究

神经网络可用来建立非线性动态系统的模型，其辨识模型可分为串联并联辨识模型和并联辨识模型两种，后者的思路源于基于参考模型自适应方案的输出误差辨识模型，对观测扰动有较强的抑制能力。本文对这种神经网络并联辨识结构的收敛性进行了研究，指出在网络参数满足一定条件时并联预测过程收敛，且并联辨识算法具有局部收敛性，仿真实验验证了上述结论。

Research on Parallel Identification Algorithm of Neural Networks

Neural networks can be used to set up models of nonlinear dynamic systems. Their identification models are classified as series-parallel model and parallel model. The latter one is developed from the output error identification model based on the reference model adaptive scheme,which has stronger capability to control observation noise. In this paper,we study the convergence of this parallel identification model and find that while the parameters of neural networks meet some prerequisites,the parallel prediction model converges and the parallel identificatin algorithm is locally convergent. Simulation results demonstrate the above conclusions.