PDF全文阅读

龙   凯¹  ,  邓经祥²  ,  李鸿钊³  ,  惠    冰³  ,  石亚强²  ,  张文俊³

 

( 1 .  济宁市公路事业发展中心 ，济宁 272008 ;  2.  葛洲坝集团交通投资有限公司 ，宜昌 443005 ;

3 .  山东省交通科学研究院 ，济南 250031)

摘   要：路基压实度与道路的质量密切相关，直接影响工程的稳定性和耐久性。为了建立路基压实度预测模型，开展路基压实度现场试验，通过控制碾压次数、碾压速度和含水率的试验方法对压实度的影响规律 ，采  用非线性回归、决策树、支持向量机 、神经网格和 XGBoost 等算法建立了6 个压实度预测模型，对其预测性能评价。结论表明：路基压实度与碾压次数呈正比，而与碾压速度呈反比。当含水率处于最优含水率时，路基压实  度最大。含水率 、碾压次数 、碾压速度对压实度的影响依次降低；支持向量机模型对于训练集预测效果较差， 决策树模型对于预测集的预测效果较差，不适用于路基压实度的预测；幂函数和对数函数两种非线性回归模型， 神经网络和 XGBoost 两种机器学习模型适用于路基压实度的预测。机器学习的预测性能要高于非线性回归模型。

关键词：路基压实度 ；现场试验 ；预测模型 ；性能评价 ；机器学习

中图分类号：U447           

文献标志码：A            

文章编号： 1005- 8249   (2025)  01- 0124- 06 

DOI:10. 19860/j.cnki.issn1005 - 8249.2025 .01 .023

 

LONG Kai ¹  ,  DENG Jingxiang² ,  LI Hongzhao³  ,  HUI Bing³  ,  SHI Yaqiang² ,  ZHANG Wenjun³

(1 . Ji ’ning Highway Devlopment Center ,  Ji ’ning 272008 ,  China;

2. Gezhouba Group Transportation Investment Co. ，Ltd. ，Yichang 443005 ,  China;

3 . Shandong Transportation Institute ,  Ji ’nan 250031 ,  China)

Abstract: Subgrade compaction is closely related to the quality of the road and directly affects the stability and durability of the project. In order to establish the prediction model of subgrade compaction, carry out the field test of subgrade compaction, through the control of the number of rolling, rolling speed and water content of the test method on the compaction of the law, the use of nonlinear regression, decision tree, support vector machine, neural lattice and XGBoost algorithms to establish a six compaction prediction model, and its prediction performance evaluation. The conclusion shows:The subgrade compaction degree is positively proportional to the number of rolling times and inversely proportional to the rolling speed. When the water content is at the optimal water content, the subgrade compaction is maximum. The effects of water content, number of rolling times and rolling speed on compaction are reduced in order;the support vector machine model has a poor prediction effect on the training set, and the decision tree model has a poor prediction effect on the prediction set, which is not applicable to the prediction of subgrade compaction; the two nonlinear regression models of the power function and the logarithmic function, and the two machine learning models of the neural network and the XGBoost are applicable to the prediction of the compaction degree of the subgrade. . The prediction performance of machine learning is higher than the nonlinear regression model.

Key words: subgrade compaction degree; field test; predictive modeling; performance evaluation; machine learning

基金项目： 山东省交通运输科技项目  (2023B45) 。

作者简介：龙   凯  (1988—) ,  男 ，硕士，工程师，研究方向：从事公路管理工作。

通信作者：李鸿钊  (1998—) ,  男 ，硕士，助理工程师 ，研究方向：路基压实。

收稿日期：2024 - 03 - 26