基于回归神经网络的非线性时变系统辨识

为克服基于前馈神经网络的非线性系统辨识算法存在需预先估计系统输入输出滞后阶数的缺陷，提出一种基于回归神经网络的非线性时变系统的辨识算法，针对现有的回归网络学习算法大多采用梯度算法，收敛速度缓慢问题，提出一种具有快速收敛性的扩展卡尔曼滤波学习算法，大大提高了学习收敛速度，并推导了一种基于单个神经元的局部化算法，减少了计算量，仿真实例证明，所提出的算法是有效的。

Identification of nonlinear time varying systems based on recurrent neural networks

A common drawback of the existing identification algorithms based on feedforward neural networks for nonlinear time varying systems is that the input and output delay orders of a system must be estimated in advance. A new identification approach based on recurrent neural networks is presented toovercomethisdrawback.The learning algorithms based on extended Kalman filter are derived. Comparedwiththetrainingalgorithmsofmost existing recurrent networks based on the gradient approach, the proposed algorithms largelyimprovethelearningconvergence,and the local algorithm reduces thecomputationcost.The simulation results demonstrate the effectiveness of the proposed approach.