基于XGBoost的堆场软土渗透系数反演研究

Inversion of permeability coefficient of soft soil in storage yard based on XGBoost

软土渗透系数的确定一直是其渗流研究的热点和难点,针对渗透系数缺乏合理确定手段的现状,以连云港堆场软土勘探项目为例,通过对静力触探数据的预处理分析选取特征参数和标签值,根据渗透系数在水平向和垂直向的差异分别调整超参数,构建了基于XGBoost机器学习算法的渗透系数反演模型。最后利用工程实测资料将XGBoost模型与传统的BP神经网络模型和前人经验公式进行对比分析,结果表明,XGBoost模型对渗透系数的预测准确度高于其它方法。

The determination of the coefficient of permeability of soft soil has always been its seepage research hot spot and the difficulty.In view of the present situation of lack of reasonable methods for determining the permeability coefficient,Lianyungang storage yard soft soil exploration project is taken as an example.Characteristic parameters and label values are selected by pre-processing analysis of data of cone penetration test,and hyperparameters are adjusted according to the difference between horizontal and vertical permeability coefficients,with a permeability coefficient inversion model based on XGBoost machine learning algorithm constructed.Finally,the XGBoost model is compared with the traditional BP neural network model and the previous empirical formulas by using the engineering measured data.The results show that the XGBoost model is more accurate than other methods in predicting the permeability coefficient.