偏最小二乘法回归在水利工程安全监测中的应用

Application of partial least-squares regression to safety monitoring of water conservancy projects

针对常规最小二乘法回归难以有效识别和消除自变量因子间的多重相关性影响这一不足,对偏最小二乘法回归在水利工程安全监测数据分析中的应用进行了研究。采用偏最小二乘法进行回归建模分析,将模型拟合与非模型式的数据内涵分析有机结合,同时实现回归建模、数据结构简化以及因子间的多重相关性分析,并通过交叉有效性检验来确保模型精度。对绕坝渗流地下水位实测资料的建模分析表明,偏最小二乘回归法能有效克服因子间的多重相关性影响,所分离出的因子变量对实测结果具有更好的物理成因解释能力,因而在水利工程安全监测及有关数据的统计分析方面具有广阔的应用前景。

Aimed at the shortages of the least-squares method regression being difficult to identify the multicollinearity of independent variable factors and eliminate its effects on the precision of regression model, the application of partial least-squares regression（PLSR） to safety monitoring of water conservancy projects was studied. By taking the measured data analysis of seepage water table around dam as an example, the PLSR was introduced to simulate and analyze the relationship between independent variables and dependent variables. The PLSR method combines the model analysis With non-model style data connotation analysis, thus, the establishment of PLSR model, simplification of data structure and the multicollinearity analysis of independent variable factors can be simultaneously carried out. The PLSR model accuracy was controlled by the cross validation test. The result of project illustration analysis indicates that the PLSR method can effectively conquer the multicollinearity disturbance of independent variable factors, and the independent variable factors separated by the PLSR model have a stronger explanation capability of physical genesis analysis on monitoring dependent variables. So, the PLSR method has a wider foreground of application than the least-squares regression method in safety monitoring data analyses and related statistic analyses of water conservancy projects.