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

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.