External deformation monitoring of high core rock-fill dams (HCRFDs) is an important and difficult part of safety monitoring. The traditional method of external deformation monitoring and data analysis for HCRFDs is to use a total station for small angle observations and establish a regression model to analyze the results. However, the small angle method has low accuracy and a low automation degree, and there is multicollinearity between the independent variables, which affects the parameter estimation and leads to the failure of model establishment. The angle forward intersection method is adopted in this paper for observation, and an improved partial least squares method (IPLS) is proposed to eliminate the multicollinearity of the independent variables. Compared to the traditional method, the improved observation method exhibits high accuracy and a high automation degree. The new data analysis method can not only eliminate multicollinearity but also improve the interpretation ability of the model. The data from the initial stage of water storage shows that the displacement increases with the increase in the upstream water level and time, and the speed of water storage is proportional to the displacement. The water level and time are the main influencing factors. This conclusion provides a theoretical basis for reservoir management departments to control water levels and gate opening and closing. The method in this paper can be applied to arch dams, gravity dams, and other types of waterpower engineering systems.
Keywords: angle forward intersection method; geodetic control network; high core rock-fill dams (HCRFD); improved partial least squares method; machine learning; total station..