平稳时序数据的Bootstrap辨识及其改进算法研究

摘要

在平稳时序数据的自回归辨识过程中,通过引入Bootstrap方法对辨识残差序列进行重采样、参数估计及模型修正等处理.对现有的基于Bootstrap方法的自回归辨识算法进行了如下的两点改进:一是以自回归模型阶数为Bootstrap方法的滑动窗口宽度,对残差序列进行重抽样处理;二是基于矩阵奇异值的迭代分解理论,对Bootstrap重抽样序列的参数进行求解.实验表明,上述改进能有效地提高原有算法的辨识精度和辨识速度.%In the process of identification of auto-regressive stationary time series data,by introducing Bootstrap method,to resampling,parameter estimation and model updating of identification residual sequence the identification accuracy can be imepoved for the original model.The following two improvements on the existing Bootstrap method are preseated based on the auto-regressive identification algorithm.One is to use auto-regressive model as sliding window width,toprocess the residuals resampling processing;The another is based on iterative matrix singular value decomposition theory to sdve the parameters of the Bootstrap sequence for re-sampling.Experimental results show that the proposed algorithm can effectively improve the identification accuracy and speed of the existing algorithms,and improve the availability of the existing algorithms.