以系数矩阵条件数的定义为基础,从数值分析的角度推导了观测值含有扰动、观测值和系数矩阵同时含有扰动以及系数矩阵含有扰动情况下的最小二乘和数据最小二乘扰动分析公式;在总体最小二乘严密求解公式的基础上,以奇异值分解为工具推导了总体最小二乘的扰动分析公式;通过分析发现,总体最小二乘扰动分析公式是最小二乘和数据最小二乘扰动分析公式的概括(统一)形式。
On the basis of the condition number definition of coefficient matrix and from the viewpoint of numerical analysis, the perturbation analysis equations of least squares (LS) and data least squares (DLS) are de- duced, in following cases when the perturbation is only contained in the observation data, the perturbations are con- tained in the observation data and coefficient matrix, and the perturbation is only contained in the coefficient ma- trix. On the basis of the closed-form expression of the total least squares (TLS) solution and with the tool of the singular value decomposition (SVD), the perturbation analysis equation of TLS is deduced. Through the analysis, the perturbation analysis equation of TLS is the unified form of the perturbation analysis equation of LS and DLS.