关键词: 弹性网络, 麻雀搜索算法, XGBoost, 时空多因子, 特征选取

Abstract: For dam deformation, some of the previous single-point models did not consider the spatial correlation of dam monitoring data and met difficulties in describing its overall response characteristics; The traditional regression models neglect the nonlinear relationship between the environmental and deformation quantities, resulting in poor prediction accuracy. To improve the prediction, this paper develops a predictive model based on an empirical modal decomposition of monitoring data by using an adaptive noise-complete set, or the technique of wavelet packet noise reduction. This model is combined with feature selection through an elastic network for the deformation factor under spatial correlation, considers the cross validation of the effectiveness of feature factors, and adopts the sparrow search algorithm extreme gradient to enhance computational efficiency. We examine the optimal factor set considering spatial correlation based on the deformation data measured at the Jinping arch dam. Comparison of the MSE and RMSE parameters of several models verifies the high accuracy and generalizability of our new method, which is useful for analysis of dam deformation patterns.

Key words: elastic net, sparrow search algorithm, XGBoost, spatiotemporal multi-factor, feature selection