基于机器学习的沥青路面压实度质量评估  

Quality Evaluation on Compaction of Asphalt Pavements Based on Machine Learning

作　　者：赵琪 张健 张智民 陈镇文 刘泽佳[1] 周立成[1] 刘逸平[1] ZHAO Qi;ZHANG Jian;ZHANG Zhimin;CHEN Zhenwen;LIU Zejia;ZHOU Licheng;LIU Yiping(School of Civil Engineering and Transportation,South China University of Technology,Guangzhou 510641,Guangdong,China;Guangzhou Chengtou Road and Bridge Engineering Co.,Ltd.,Guangzhou 510199,Guangdong,China)

机构地区：[1]华南理工大学土木与交通学院,广东广州510641 [2]广州城投路桥工程有限公司,广东广州510199

出　　处：《济南大学学报（自然科学版）》2024年第3期331-340,共10页Journal of University of Jinan(Science and Technology)

基　　金：国家自然科学基金项目(11972162);广东省自然科学基金项目(2023A1515012942)。

摘　　要：为了提高沥青路面压实度预测的准确率,基于机器学习对沥青路面进行压实度质量评估;通过对比实际工程中不同压路机在沥青路面上的振动规律,构建振动压路机-沥青路面系统动力学模型,利用Simulink数值仿真软件对所构建的模型进行仿真,设计8种工况,对比压实度计值、压实度控制值、机械驱动功率、滚轮综合刚度、填筑体能量、单位体积压实功率6个压实度质量评估指标在各工况下的适用性差异;采用支持向量机、逻辑回归、k最近邻、决策树、朴素贝叶斯法5种传统的机器学习方法对各工况下的压实度质量评估指标样本进行训练,对比压实度预测准确率;设计压路机在不同碾压次数时的碾压路线,对比分别采用最优压实度质量评估指标和单一压实度质量评估指标的压实度预测准确率。结果表明:不同压实度质量评估指标在不同工况下的适用性不同,即使在同一种工况下,不同碾压次数时的适用性也存在差异;采用最优压实度质量评估指标代替单一压实度质量评估指标,压实度预测准确率提高5.8%;在5种传统的机器学习方法中,朴素贝叶斯法预测最优压实度质量评估指标类型的准确率最高,为96.22%。To improve accuracy of asphalt pavement compaction prediction,quality evaluation on compaction asphalt pavements was conducted on the basis of machine learning.By comparing vibration patterns of different rollers on asphalt pavements in practical engineering,a dynamical model of vibration roller-asphalt pavement system was constructed.The constructed model was simulated by using Simulink numerical simulation software,and eight working conditions were designed to compare applicability differences of six quality evaluation indicators of compaction under different working conditions including compaction meter value,compaction control value,machine drive power,roller-integrated stiffness,filling energy,and unit volume compaction power.Five traditional machine learning methods,namely support vector machine,logistic regression,k-nearest neighbor,decision tree,and naive Bayes,were used to train quality evaluation indicator samples of compaction under different working conditions,and prediction accuracy of compaction was compared.Rolling routes of the roller with different rolling times were designed to compare predict accuracies of compaction using the optimal quality evaluation indicators of compaction and the single quality evaluation indicators of compaction.The results show that the applicability of different quality evaluation indicators of compaction is different under each working condition.Even under the same working condition,the applicability with different rolling times is also different.Using the optimal quality evaluation indicators of compaction to evaluate indicators instead of the single quality evaluation indicators of compaction,the prediction accuracy of compaction is increased by 5.8%.Among the five traditional machine learning methods,naive Bayes method has the highest accuracy of predicting types of the optimal compaction quality evaluation indicators,which is 96.22%.

关 键 词：沥青路面 压实度质量评估 机器学习 压实度 动力学模型 

分 类 号：U416.217[交通运输工程—道路与铁道工程]