非线性动态传感器系统的H模型神经网络辨识

针对非线性动态传感器模型辨识问题，提出一种新的Hammerstein模型神经网络结构辨识法。非线性动态传感器系统采用Hammerstein模型描述，将系统分解为非线性静态增益串接线性动态环节。再设计一种网络权系数对应于相应的Hammerstein模型参数的新型神经网络结构，推导了基于反向传播的网络权系数调整方法。通过网络迭代训练同时得到静态与动态两个环节的模型参数。最后通过一个H模型的数值仿真来验证方法的有效性，仿真结果表明所提辨识方法是有效的。

Identification of Transducer Nonlinear Dynamic System Using Hammerstein Neural Network

For identification nonlinear dynamic model of transducer, a novel Hammerstein neural network structure is presented. Firstly, the nonlinear dynamic system is described by a Hammerstein model which consists of a nonlinear static gain in cascade with a linear dynamic part. Secondly, a novel neural network structure is designed, in which the weights are corresponding with the parameters of the Hammerstein model, backward-propagation methods for the adjustment of weights in the network are discussed. Finally, parameters of the nonlinear static gain part and the linear dynamic in the Hammerstein model are determined simultaneously by iterative training. A numerical simulation of Hammerstein model is provided to validate the effectiveness. Simulation results show that the suggested identification schemes are practically feasible.