摘要
根据递阶辨识原理,研究了类多变量方程误差系统和类多变量方程误差ARMA系统递阶随机梯度方法和递阶梯度迭代方法、递阶最小二乘方法和递阶最小二乘迭代方法.进一步利用多新息辨识理论,推导了递阶多新息梯度辨识方法和递阶多新息最小二乘辨识方法.为减小计算量,推导了基于滤波的类多变量方程误差ARMA系统递阶辨识方法和递阶多新息辨识方法.讨论了几个典型辨识算法的计算量,并给出了计算参数估计的步骤.
According to the hierarchical identification principle,this paper presents the hierarchical stochastic gra-dient algorithms and the hierarchical gradient based iterative algorithms, the hierarchical least squares algorithms and the hierarchical least squares based iterative algorithms for multivariable equation-error-like systems and multi-variable equation-error ARMA-like systems,and further derives the hierarchical multi-innovation gradient algorithms and the hierarchical multi-innovation least squares algorithms. In order to reduce computational burdens,this paper derives the filtering based hierarchical identification algorithms and the filtering based hierarchical multi-innovation identification algorithms for multivariable equation-error ARMA-like systems using the filtering technique. Finally, the computational efficiency and the computational steps of some typical identification algorithms are discussed.
出处
《南京信息工程大学学报（自然科学版）》
CAS
2014年第5期385-404,共20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
最小二乘搜索
多新息辨识理论
递阶辨识原理
类多变量系统
数据滤波技术
parameter estimation
recursive identification
gradient search
least squares search
multi-innovation identification theory
hierarchical identification principle
multivariable-like system
data filtering technique
