基于神经网络的Hammerstein OE非线性系统参数估计

Parameter Estimation of Hammerstein OE Nonlinear System Based on Neural Network

针对噪声干扰下的Hammerstein非线性动态系统,提出一种基于神经网络的Hammerstein OE(Hammer-stein Output Error)非线性系统参数估计方法。在该系统中,利用BP神经网络和自回归模型分别建立静态非线性模块和动态线性模块,并提出两阶段参数估计方法:第一阶段,利用辅助模型递推最小二乘方法估计动态线性模块的参数,解决了系统中间变量不可测问题;第二阶段,为了改善参数学习收敛率,利用含有动量项的随机梯度下降方法估计BP神经网络的权值。仿真结果表明,提出的方法能够有效估计Hammerstein OE非线性系统参数。

For Hammerstein nonlinear dynamic systems corrupted by noise,a parameter estimation method neural network based Hammerstein OE nonlinear system is proposed.In this system,BP neural network and autoregressive model is used to build static nonlinear block and dynamic linear block respectively,and two stage parameter estimation scheme are proposed.In the first stage,the parameters of dynamic linear block are estimated by using auxiliary model recursive least square method,which solves the problem that the intermediate variables of the system cannot be measured.In the second stage,in order to improve the learning convergence rate,the random gradient descent method with momentum term is used to estimate the weights of BP neural network.Simulation results show that the proposed parameter estimation method can effectively estimate the parameters of Hammerstein OE nonlinear system.