SSA-XGBoost与时空特征选取的大坝变形预测模型

Dam deformation prediction model selected by SSA-XGBoost with temporal and spatial features

针对目前部分单测点模型未考虑大坝监测数据空间关联性、难以描述大坝变形整体响应特性的问题,以及传统回归模型未考虑环境量与变形量的非线性关系导致预测精度较低的问题,本文提出了一种预测模型,包括对监测数据进行基于自适应噪声完备集合经验模态分解-小波包降噪,结合弹性网络对考虑了空间关联性的变形效应量因子进行特征选取,辅以交叉验证特征因子的有效性,并使用麻雀搜索算法提高计算效率。基于锦屏一级拱坝实测变形数据,探究了考虑空间关联性的最优因子集,并通过对比多种模型的MSE、RMSE等参数验证了本文方法的有效性,在大坝变形性态分析中具有一定应用价值。

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.