变异函数的稳健性及特异值处理方法的研究

RESEARCH ON THE ROBUSTNESS OF VARIABLE DIFFERENCE FUNCTION AND THE TREATMENT OF THE CUTLIER

本研究在分析变异函数不稳健性产生的机理基础上 ,对正态变换的几种方法及识别和剔除特异值的方法都做了初步尝试 ,并用“估计邻域法”识别并处理特异值取得了较好的稳健效果。在此基础上 ,用“中位数中位偏差法”、“中位调节法”、“位置的尤伯估计”、“α切尾均值法”等 4种稳健统计学方法较系统地对水盐采样数据作了处理和比较研究。处理后得到的理论变异函数图中拱高值 ( C0 +C )与样本方差σ2 的比较结果经交叉检验法和估计方差检验法检验得出 :“中位数中位偏差法”处理后的模型最优。

Based on the analysis of the mechanism resueting in the unrobustness of the uariable Difference Funotion,the rescarch makes the first affemptin experimenting with the means of Normal Uariation and the means of outlier Recognition and Rejection.Better resultes were achievea when the Neighboring Domain Assumption Method was used to recognize and treat the outlier.Based on what have been done,Sampling data were systematically treated by four Robust statistical methods,Middle deuiation of Median Method,Median Regulation Method,Huber's estimation Method of location and α-fail method Besides,Comparison Study was done in every treatment.According to the comparision resule of the resulting (C 0+C)theoretical diagram of variable Difference Function,ana sample σ 2,together with Crossing Examination method and Assumption Examination Method,the best formulae resulted from Miadle deriation of Median Method.