Deformation prediction model of high-core rockfill dam based on EBWO-LightGBM

Abstract

The deformation of high-core rockfill dams is one of the most direct and reliable indicators for assessing the operational state and safety conditions of dams. Accurately predicting dam deformation and uncovering its patterns are crucial for adjusting construction plans or improving dam safety management strategies. To address the issues of local optima entrapment and handling large-scale data in dam deformation prediction models, the Enhanced Beluga Whale Optimization(EBWO) algorithm is utilized to optimize the hyperparameters of the Light Gradient Boosting Machine(LightGBM), aiming to obtain the optimal hyperparameter combination. This forms the basis for constructing a deformation prediction model of high-core rockfill dams based on EBWO-LightGBM. Taking a high-core rockfill dam in the Southwest as a case study, EBWO-LightGBM, standard LightGBM, ELM, SVM, and RF models were developed, and their predictions analyzed. The EBWO-LightGBM prediction model achieved a correlation coefficient(R²) of 98.2%, an improvement of 3.81% over standard LightGBM, and a reduction in Root Mean Square Error(RMSE) of 11.72%. This demonstrates that the model effectively balances global and local performance, enhancing predictive capabilities. Compared to ELM, SVM, and RF models, the EBWO-LightGBM model showed an increase in R² by 4.80%, 4.58%, and 4.25% respectively, and a decrease in RMSE by 13.41%, 13.18%, and 12.75%, confirming its superiority.

Keywords

high-core rockfill dam / deformation prediction model / EBWO / LightGBM