对最小二乘估计作线性变换,使得新估计是真值的最优拟合,并且包含模型参数的先验误差协方差阵。从滤波因子的角度对正则化方法进行统一,提供了常见的Tikhonov正则化方法、截断奇异值法、广义岭回归方法等的滤波因子与对应的误差协方差阵的特征值。
In this paper,the estimator is based on a linear transformation of the LS estimation.It provides an optimal solution to the true and a constrained covariance matrix.The regularization techniques are unified with the respect to filtering factors.Filtering factors and corresponding eigenvalue of error covariance matrix of some known regularization estimators as Tikhonov-Philips regularization,truncated singular value decomposition,generalized ridge regression and others are provided.