Prediction of earthquake damage of reservoir dam based on RS-XGBoost

The destruction of hydraulic structures caused by earthquakes brings great risks to the society. It is of great significance to quickly predict the post-earthquake structural damage of reservoir dams for pre-disaster prevention and post-disaster rapid decision-making. In order to quickly and accurately predict the possible earthquake damage of dams, this paper collected 189 real earthquake damage data of reservoir dams in Tonghai earthquake in 1970 and Wenchuan earthquake in 2008, and selected 10 input parameters of earthquake intensity and dam structure as the main features, and the output results were the earthquake damage grades of cracks, leakage and landslides to construct samples. The sample is divided into 80% training set, 10% verification set and 10% test set, and the XGBoosting algorithm is optimized by four super-parameter optimization algorithms: random search, grid search, bayesian optimization and particle swarm optimization. The results show that the RS-XGBoosting model has high data adaptability and prediction accuracy. The training set determination coefficients R2 of the three earthquake damage phenomena are as high as 0.99, 0.98 and 0.92, and the verification set determination coefficients R2 are 0.93, 0.84 and 0.82. The model can provide a reference method for earthquake damage prediction of reservoir dams.