顾及设计矩阵误差的AR模型新解法

A New Method to Solving AR Model Parameters Considering Random Errors of Design Matrix

在自回归模型求解中,设计矩阵和观测值均存在误差,传统的最小二乘法不能很好地解决这一问题。本文提出了一种顾及设计矩阵误差的AR模型新解法,通过引入虚拟观测值,使观测向量与设计矩阵不仅同源而且带误差的元素个数相同,然后通过对观测方程进行等价变换巧妙实现了在最小二乘框架下求解自回归问题。利用模拟数据及实测数据分别对新算法进行了内符合精度检验,并利用实测数据对新算法进行外符合精度检验,结果表明新算法得到的结果显著优于奇异值分解(singular value decomposition,SVD)解法及传统最小二乘解法,验证了算法的精度和有效性。

The ordinary least square method could not solve the problem that the error exist both in design matrix and observation vector while compute parameter values of AR model.In this article,a new method is proposed which consider the random errors of design matrix.The source of design matrix and observation vector is same and the amount of parameters contain error can be equal by introducing virtual observation.Then,this problem could be solved under the framework of normal least square by equivalence transformation of observation equation.The result of this new method is superior to SVD method and normal least square method by simulation date and observation data which verify the feasibility and effectiveness of this method.