大坝安全监控模型因子相关性及不确定性研究

针对最小二乘法难以克服因子多重共线性对回归模型精度影响的不足，本文对大坝安全监控模型因子间的相关性及其不确定性进行了研究。引进偏最小二乘法，对大坝安全监测变量及其影响因子进行偏最小二乘回归分析，将建模预测分析方法与非模型式的数据内涵分析有机结合，可同时实现回归建模、数据结构简化以及因子相关的不确定性分析，所建立的大坝安全监控模型，其精度可通过交叉有效性检验来控制。工程应用实例和模型对比分析研究表明，偏最小二乘回归模型能有效克服各类因子变量间的多重共线性对模型拟合精度及其预测能力的影响，因而比目前常用的最小二乘回归模型更具广泛适用性。

Multiple co-linearity and uncertainty of factors in dam safety monitoring model

The partial least-squares regression (PLSR) method is introduced to analyze the variable of dam safety monitoring model and the factors affecting these variables. This method combines the forecasting by means of model with non-model style data connotation analysis, thus, the establishment of model by mean of regression, simplification of data structure and the analysis of correlative uncertainty of model factors can be carried out simultaneously. The accuracy of the dam safety monitoring model is controlled by the cross validation test. The comparison of in situ monitoring data with model forecasting shows that the proposed method can effectively overcome the effects of multiple co-linearity among variables on fitting accuracy and forecasting capability of the safety monitoring model.