利用随机森林算法建立沥青路面车辙预测模型  

Using random forest algorithm to build rutting prediction model of asphalt pavement

作　　者：许有成 杜亚冰 XU You-cheng;DU Ya-bing(School of Civil Engineering&Architecture,Anhui University of Science&Technology,Huainan 232001,China;School of Mathematics&Physics,Henan University of Urban Construction,Pingdingshan 467036,China)

机构地区：[1]安徽理工大学土木建筑学院,安徽淮南232001 [2]河南城建学院数理学院,河南平顶山467036

出　　处：《河南城建学院学报》2022年第2期43-48,共6页Journal of Henan University of Urban Construction

基　　金：河南省科技攻关计划项目(212102210172)。

摘　　要：RFR(随机森林回归)是一种非常经典的机器学习算法,可用于建立路面车辙破坏模型以提高其预测准确度,并将其与传统的线性回归模型进行比较。基于Python编程语言环境下的Sikit-learn扩展包中随机森林模块进行模型搭建,训练模型数据来源于NCRHP(国家合作公路研究计划)01-37A项目生成的两份报告和LTPP(长期路面性能)网站的记录。结果表明:RFR模型明显优于线性回归模型,在测试集中测量值和预测值之间的相关系数R2值从0.392增大到0.875,标准差Se从3.23 mm下降到1.41 mm。因此,RFR模型能够更准确地预测沥青路面车辙。RFR(Random Forest Regression)is a classic machine learning algorithm,which can be used to build road rutting failure model to improve its prediction accuracy,and compare it with the traditional linear regression model.The model is based on the random forest module in Sikit-learn expansion package in Python programming language environment.The training model data comes from two reports generated by NCRHP(National Cooperative Highway Research Program)01-37A project and the records of LTPP(Long-term Pavement Performance)website.The results show that the RFR model is obviously superior to the linear regression model.The correlation coefficient R2 between the measured value and the predicted value in the test set increases from 0.392 to 0.875,and the standard deviation Se decreases from 3.23 mm to 1.41 mm.Therefore,RFR model can predict asphalt pavement rutting more accurately.

关 键 词：MEPDG 车辙预测 决策树 随机森林 线性回归 

分 类 号：U416.217[交通运输工程—道路与铁道工程]