PDF全文阅读

        戴   汀1 ,2 ,  李向国1 ,2 ,  王    凯3

 

( 1 .  道路与铁道工程安全保障省部共建教育部重点实验室（石家庄铁道大学），石家庄 050043 ;   2.  石家庄铁道大学 土木工程学院 ，石家庄 050043 ;  3 .   中建三局一公司基础分公司 ，武汉 430064)

摘   要 ：针对传统方法难以快速精准预测陇中地区黄土湿陷性的问题 ，开展基于支持向量机回归的黄土湿 陷性预测研究。研究依托陇中地区某铁路项目 ，于人工探井中采集样土进行室内土工试验 ， 以获取土的相关物 性指标数据。通文献分析法和数理分析 ，探究黄土湿陷系数与各物性指标的相关性。在此基础上 ，基于支持 向量机 ，采用高斯核函数构建黄土湿陷性预测模型。以 45 组试验数据作为训练集 ，9 组作为测试集进行误差分 析。结果表明：预测模型训练集平均绝对误差为 0. 003 、平均偏差误差为 0. 001 、均方根误差为 0. 015 9 ;  测试集 平均绝对误差为 0. 013 2 、平均偏差误差为 0. 001 8 、均方根误差为 0. 015 ,  表明该模型预测精度良好 ，可有效预 测陇中地区黄土湿陷系数 ，为实际工程中黄土湿陷性的预测提供了可靠的新方法。

关键词：黄土湿陷性 ；物性指标 ；相关性 ；支持向量机 ；预测模型

中图分类号：TU444           文献标志码：A   

         

文章编号： 1005- 8249   (2025)  03- 0073- 05 

DOI:10. 19860/j.cnki.issn1005 - 8249.2025 .03 .014

DAI Ting1 , 2  ,  LI Xiangguo1 , 2  ,  WANG Kai3

(1. Key Laboratory of Road and Railway Engineering Safety and security, Ministry of Education (Shijiazhuang Railway University), Shijiazhuang 050043, China;

2. School of Civil Engineering, Shijiazhuang University of Railways, Shijiazhuang 050043, China

3. Basic Branch of China Construction Third Bureau One Company, Wuhan 430064, Hubei)

 

Abstract:   To address the challenges in rapid and accurate prediction of loess collapsibility in Central Gansu using conventional methods, a support vector machine regression-based predictive model was developed. Leveraging data from a railway project in this region, soil samples collected from manually excavated test pits underwent comprehensive laboratory geotechnical testing to obtain key physical property indices.Correlations between loess collapse coefficients and physical indices were systematically investigated through literature analysis and mathematical statistics. A Gaussian kernel function-based SVM prediction model was subsequently established. The model was trained with 45 experimental datasets, while 15 datasets were utilized for validation and error analysis. Results demonstrate that the model achieves a mean absolute error (MAE) of 0.003, mean bias error (MBE) of 0.001, and root mean square error (RMSE) of 0.0159 on the training set. For the testing set, performance metrics indicate MAE=0.0132, MBE=0.0018, and RMSE=0.015. These findings confirm the model's high predictive accuracy for estimating loess collapse coefficients in Central Gansu, providing a reliable novel methodology for practical engineering applications.

Key words:   loess collapsibility ; physical property index ; correlation ; support vector machine ; prediction model

基金项目： 国家自然科学基金  (52378453) 。

作者简介：戴   汀  (1998—) ,  男 ，硕士研究生 ，研究方向： 黄土湿陷性。

通信作者：李向国  (1973—) ,  男 ，博士 ，教授 ，研究方向：铁路线路结构设计 、施工等方面。

收稿日期：2025 - 01 - 18