摘要
针对多元线性回归系统,利用耦合辨识概念和多新息辨识理论,讨论了多元随机梯度算法、多元多新息随机梯度算法,以及变递推间隔多元多新息梯度算法,进一步分解多元系统为一些子系统,给出了耦合子系统随机梯度算法、耦合随机梯度算法、耦合子系统多新息随机梯度算法、耦合多新息随机梯度算法,并将这些方法推广到多元伪线性滑动平均系统和多元伪线性自回归滑动平均系统.文中给出了几个典型耦合随机梯度算法、耦合多新息随机梯度算法的计算步骤和示意图.
For multivariate linear regression systems,using the coupling identification concept and the multi-inno-vation identification theory,this paper discusses a multivariate stochastic gradient algorithm,a multivariate multi-in-novation stochastic gradient algorithm,and an interval-varying multivariate multi-innovation gradient algorithm,de-composes a multivariate system into several subsystems,and presents a coupled subsystem stochastic gradient algo-rithm,a coupled stochastic gradient algorithm,a coupled subsystems multi-innovation stochastic gradient algorithm and a coupled multi-innovation stochastic gradient algorithm.These methods are extended to multivariate pseudo-lin-ear moving average systems and multivariate pseudo-linear autoregressive moving average systems.Finally,this paper gives the steps and diagrams for computing the parameter estimates using several typical coupled stochastic gradient algorithms and coupled multi-innovation stochastic gradient algorithms.
出处
《南京信息工程大学学报（自然科学版）》
CAS
2014年第1期1-16,共16页
Journal of Nanjing University of Information Science & Technology（Natural Science Edition）
基金
国家自然科学基金(61273194)
江苏省自然科学基金(BK2012549)
高等学校学科创新引智"111计划"(B12018)
关键词
参数估计
递推辨识
梯度搜索
最小二乘
辅助模型辨识思想
多新息辨识理论
递阶辨识原理
耦合辨识概念
多元系统
parameter estimation
recursive identification
gradient search
least squares
auxiliary model identifica-tion idea
multi-innovation identification theory
hierarchical identification principle
coupling identification concept
multivariate system
