多元伪线性回归系统部分耦合多新息随机梯度类辨识方法

摘要

For multivariate pseudo-linear regressive moving average systems,a multivariate extended stochastic gra-dient(ESG) algorithm is discussed.In order to reduce the computational cost of the identification algorithm,we de-compose a multivariate system into several subsystems,and derive a partially coupled(subsystem) ESG algorithm and a partially coupled( subsystem) multi-innovation ESG algorithm according to the coupling identification concept and the multi-innovation identification theory. Furthermore, we extend these methods to multivariate pseudo-linear autoregressive moving average systems and present a partially coupled( subsystem) generalized extended stochastic gradient ( GESG ) algorithm and a partially coupled ( subsystem ) multi-innovation GESG algorithm. The computational efficiencies of the multivariate ESG algorithm,the partially coupled ESG algorithm and the partially coupled multi-innovation ESG algorithm are analyzed.%针对多元伪线性滑动平均系统，讨论了多元增广随机梯度算法，为减小算法的计算量，将系统分解为一些子系统，给出了子系统增广随机梯度算法，利用耦合辨识概念和多新息辨识理论，推导了部分耦合（子系统）增广随机梯度算法、部分耦合（子系统）多新息增广随机梯度算法。进一步将提出的方法推广到多元伪线性自回归滑动平均系统，给出了部分耦合（子系统）广义增广随机梯度算法、部分耦合（子系统）多新息广义增广随机梯度算法。文中分析了多元增广随机梯度算法、部分耦合增广随机梯度算法、部分耦合多新息增广随机梯度算法的计算量。