基于时变NARMA模型的非线性时变系统辨识

在非线性自回归滑动平均模型NARMA(NonlinearAutoRegressiveMovingAverage)中引入时间变量,将其扩展为时变NARMA模型,用Taylor展开将模型中的非线性函数展开为关于输入输出的多项式,得到关于参数线性时变的多项式形式的时变NARMA模型,再用基序列拟合模型的时变参数得到关于参数线性时不变的模型,最后用递推最小二乘法估计模型参数。仿真算例证明,与小波网络方法相比,辨识精度高,计算量小。

NONLINEAR TIME-VARYING SYSTEM IDENTIFICATION BASED ON TIME-VARYING NARMA MODEL

Introducing time variable into the NARMA (Nonlinear Auto Regressive Moving Average) model make it expand to time-varying NARMA model. The nonlinear function of the model can be expanded to a polynomial of input and output using Taylor expansion, and the polynomial time-varying NARMA model that is linear to the parameters is obtained. Using base sequences to fit the time-varying parameters of the model, the nonlinear time-varying system is transformed into a time-invariant linear one, the parameters of which can be estimated by recursive least square algorithm. The results of simulation examples show that the identification accuracy and the computational complexity of this method are better than those of wavelet neural network method.