逐步回归和偏最小二乘回归模型在混凝土重力坝变形监测中的应用

Application of Stepwise Regression and Partial Least Squares Method in Gravity Dam Deformation Monitoring

针对重力坝变形监测中单一回归模型存在的不足,引入水位、温度和时效各因子,分别建立了逐步回归模型和偏最小二乘回归模型,选取A电站14年的资料共88组样本点,分别采用两种模型对上下游方向S02单测点进行分析。结果表明,两种模型在很大程度上对混凝土重力坝变形监测结果分析均适用,实测值与逐步回归计算值和偏最小二乘回归计算值基本吻合,但偏最小二乘回归模型各水位、温度和时效各因子有效避免了舍弃线性相关自变量,对大坝变形的影响较逐步回归模型更加接近实际规律,在大坝变形监测中将二者方法得到的结果结合起来,得到的监测资料更为准确。

Aiming at the shortcomings of single regression model for application of gravity dam monitoring,the stepwise regression and partial least squares regression model are established by introducing water level,temperature and aging factors.Choosing 88 groups of sample points for 14 years data of power station A,two models are used to analyze the single observation point S02 in direction of upstream and downstream.The results show that two models apply to analysis of concrete gravity dam deformation monitoring in great extent;the measured values are basically consistent with the values obtained by stepwise regression and partial least-squares regression;but water level,temperature and aging factors in partial least squares regression model overcome the impact of linear correlation independent variables effectively,and the influence on the dam deformation is more close to the actual situation.Therefore,combination of stepwise regression and partial least squares regression is more accurate for getting dam deformation monitoring values.