多变量自回归模型的快速辨识方法

本文提出了多变量AR(Autoregressive)模型建模的快速算法,该算法采用解超定矩阵的最小二乘方法,并利用正交变换技术,从而避免了最小二乘估计中矩阵求逆的病态问题,保证了数值计算的稳定性。还介绍了一种可转化为单变量建模的多变量AR模型,它可以直接利用单变量建模模块。一般情况下,可将序列进行典则分解。最后简要列出了有关结论。

The Fast Algorithm of Modelling the Multivariable AR Process

In this paper, the fast algorithm of structure determination and parameter estimation of multivariable AR model is presented. This algorithm consists in the least square method for solving contradictory matrice equations. By this algorithm the orthogonal transformation technique is used to avoid stiff problems of matrice calculations, ensuring the stability of numerical calculation. Also introduced in this paper is a kind of multivariable AR model which can be changed into single variable to modelling,it can directly use the programs of modelling single variable AR process. Generally speaking, multivariable time series may be transformed into canonical form.A part of the conclusions are given.