多新息随机梯度辨识方法的收敛性研究

Convergence of Multi-innovation Stochastic Gradient Identification Methods

从理论上给出了多新息辨识方法的推导过程,提出了多新息随机梯度辨识方法,并运用随机过程理论分析了多新息随机梯度辨识方法的均方收敛性,给出参数估计误差上界的计算公式。分析表明数据的平稳性可以提高参数估计精度。最后给出了多新息辨识方法的各种典型变形。

The multi-innovation identification algorithm is derived mathematically,and the multi-innovation stochastic gradient identification algorithm is presented.The mean square convergence is analyzed by using the stochastic process theory,and the upper bound of the parameter estimation error is obtained.The analysis indicates that the data stationary can improve the estimation accuracy,and then the changing forms of multi-innovation identification are given.