应用随机森林模型识别南宁地铁线路地层结构特征

Research on Recognition of Stratum Structure Using Random Forest Model

摘要: 利用传统空间插值方法对钻孔数据进行三维地层建模易受主观因素的影响，为了克服地层建模过程中产生的误差，本文基于随机森林模型提出了一种新型的地层识别方法。选取南宁地铁1号线部分工程勘察资料进行研究，利用网格搜索确定模型参数后，构建随机森林模型，并与支持向量机进行对比分析。通过研究表明，随机森林模型的预测精度达到81.7%，其超参稳定性明显高于支持向量机，且预测精度受其主要参数(树的数量和最小叶子节点数)变化的影响较小；交叉验证评估结果也证实了随机森林的泛化性能更好。当样本数量较少时，随机森林模型不论是在分类精度还是稳定性方面均较好。该方法在三维地质建模中具有良好的应用价值。

Abstract: Using traditional spatial interpolation method based on borehole data for 3D stratum modeling is easily affected by subjective factors. In order to overcome the errors in the stratum modeling process, this paper proposes a new stratum recognition method based on the Random Forest model. Part of the engineering investigation data of Nanning Metro Line 1 is selected to carry out the research. After the classifier parameters are determined by the grid parameter search, the Random Forest model is constructed and compared with the Support Vector Machine. Research shows that the prediction accuracy of the random forest model reaches 81.7%, and its hyperpa-rameter stability is significantly higher than that of the support vector machine, and the predic-tion accuracy is less affected by changes in its main parameters (the number of trees and the minimum number of leaf nodes). The cross-validation results also confirm that the generalization performance of random forest is better. When the number of samples is small, the random forest classifier still has a good performance in terms of classification accuracy and stability. This method has a good application prospect in 3D geological modeling.