非线性系统的动态神经网络自适应辨识

提出了用双层动态神经网络在线辨识非线性动态系统的方法。神经网络的权重在线学习，不需要离线训练。在无逼近误差和扰动的理想情况下，所提出的在线算法能保证辨识误差趋于零，基函数持续激励条件能保证权重趋于零。在非理想情况下，权重调整律采用ｅ修正权重算法，它是ＢＰ算法的推广，不需要基函数的持续激励条件。基于李雅普诺夫稳定性理论保证了自适应辨识系统的稳定性。仿真算例说明了所提出的动态神经网络自适应辨识的有效性

Adaptive Identification Using Recurrent Neural Network for Non-linear Systems

An adaptive identification method is proposed for non linear dynamical systems using two layer recurrent neural networks. On line weight tuning algorithm with e modification is designed without the need of off line training phase. Stability analysis of the adaptive identification systems is presented. The proposed algorithm guarantees that the identification error and the neural weight estimation error are uniformly ultimately bounded for systems with disturb and neural network approximation error. A simulation example illustrates the effeciency of the proposed dynamical identification method.