Renmin Zhujiang (Jan 2024)
Comparative Study on Prediction Models for Crack Opening Degree in Concrete Dam
Abstract
In recent years, classical statistical models and machine learning models have been developed in parallel in the field of dam safety monitoring. However, there are some deficiencies in the predictive power of the former and the theoretical explanation of the latter. In this study, multiple linear regression, stepwise regression, and random forest algorithm were used to establish models for the crack opening degree of a concrete gravity dam based on the monitoring data of the crack opening degree of the concrete gravity dam. The results show that three models for predicting crack opening degree are successfully established based on the crack opening degree dataset measured in 2022. The random forest model has the best predictive ability (determination coefficient (R2) is 0.995; root mean square error (RMSE) and mean absolute error (MAE) are 0.174 mm and 0.124 mm, respectively), followed by the stepwise regression model (R2 is 0.989; RMSE and MAE are 0.192 mm and 0.151 mm). Three models both indicate that the temperature component is the main factor affecting the crack opening degree of the concrete gravity dam; by decomposing the multiple linear regression model item by item, the variation patterns of crack opening degree of the concrete gravity dam, temperature component, hydraulic pressure component, and time component are obtained. This study can provide a reference for the operation and management of the concrete gravity dam and the construction of the forecasting, early warning, drilling, and emergency plan (FEDE) platform, with a relatively high theoretical and practical significance.
