针对大坝安全监测统计模型中异方差产生的原因和可能带来的后果，用图示法及格莱斯尔检验法检验异方差性的方法，并采用方差的稳定变换及加权最小二乘估计法对存在异方差问题的统计模型进行了改进，使之降低了异方差性对模型参数估计的影响，使回归模型更好地拟合观测数据，提高了利用模型进行预测和控制的可靠度。

Aiming at the causes generating differential deviation and possible potential consequences in the statistics model of dam safety monitoring, two methods used for detecting differential deviation,i.e.,graphic method and Glejser method are presented. The improvement to the statistics model causing differential deviation is performed by using stable transformation method and weighted regression estimation method so as to decrease the influence of differential deviation on model parameters estimated values,and so the regression model can better fit observed date and raise the reliability of predication and control.