有色噪声干扰下Hammerstein非线性模型辨识

Identification of Hammerstein nonlinear model with colored noise

针对非线性Hammerstein模型存在滑动平均噪声干扰问题,提出了一种智能分离辨识方法.利用多项式模型和状态空间模型分别建立Hammerstein模型的非线性子系统和线性子系统,并设计组合信号分离辨识Hammerstein模型的静态非线性子系统和动态线性子系统.首先,分析了二进制信号不激发非线性子系统特性,推导出基于辅助变量的增广最小二乘方法辨识动态线性子系统和噪声模型的参数,有效削弱了滑动平均噪声的干扰.其次,为了提高随机梯度辨识方法的辨识精度和收敛速度,推导了遗忘因子增广随机梯度方法,基于随机信号的输入/输出数据辨识静态非线性子系统的参数.仿真结果表明,本文推导的方法能够辨识滑动平均噪声干扰下Hammerstein模型,与递推增广最小二乘方法、迭代梯度方法和辅助模型递推增广最小二乘方法相比,提出的方法能够取得较高的辨识精度.

Considering the issue that nonlinear Hammerstein models exist moving average noise,an intelligent separation identification technique is proposed.The nonlinear subsystem and linear subsystem of Hammerstein model are established by polynomial model and state space model,respectively,and the static nonlinear subsystem and dynamic linear subsystem are separately identified by utilizing designed combined signal.Firstly,analyzing the characteristic that binary signals do not excite nonlinear subsystem,the dynamic linear subsystem and the noise model parameters are identified utilizing auxiliary variables based recursive extended least squares algorithm,which can weaken the interference of moving average noise effectively.Furthermore,in order to improve the identification accuracy and convergence speed of the stochastic gradient identification algorithm,the forgetting factor extended stochastic gradient method is derived.The parameters of the static nonlinear subsystem are identified based on the input-output data of random signal.The simulation results indicate that the derived method can identify the Hammerstein nonlinear model under moving average noise interference.Compared with the recursive extended least squares method,the Gradient-based Iterative method and the auxiliary model based recursive extended least squares method,the proposed method can achieve higher identification accuracy.