基于历史通行数据的假冒绿通车逃费行为预测  

Prediction of Fake Toll-free Vehicles Based on Historical Traffic Data

作　　者：刘昱岗[1,2,4] 郑帅 徐旭东 王添碧 叶劲松[3] LIU Yu-gang;ZHENG Shuai;XU Xu-dong;WANG Tian-bi;YE Jin-song(School of Transport and Logistics,Southwest Jiaotong University,Chengdu Sichuan 610031,China;National United Engineering Laboratory of Integrated and Intelligent Transport,Southwest Jiaotong University,Chengdu Sichuan 610031,China;China Academy of Transportation Science,Beijing 100088,China;Institute of Transport Development Strategy and Planning of Sichuan Province,Chengdu Sichuan 610001,China)

机构地区：[1]西南交通大学交通运输与物流学院,四川成都610031 [2]西南交通大学综合交通运输智能化国家地方联合工程实验室,四川成都610031 [3]交通运输部科学研究院,北京100088 [4]四川省交通运输发展战略和规划科学研究院,四川成都610001

出　　处：《公路交通科技》2021年第4期92-102,141,共12页Journal of Highway and Transportation Research and Development

基　　金：国家自然科学基金项目(51774241);四川省青年科技创新研究团队项目(2020JDTD0027);综合交通运输大数据应用技术交通运输行业重点实验室开放课题项目(2018B1201)。

摘　　要：目前高速公路"绿色通道"政策全面实施,但检测方式相对滞后,货车司机容易受通行费减免诱惑假冒绿通车逃费,给运营管理单位造成巨大损失。为提高绿通车检查的准确率和效率,基于四川省高速公路联网收费系统采集的历史绿通车通行数据集,建立了针对假冒绿通车逃费行为的预测模型。首先利用数据挖掘技术,对通行数据集的属性按照重要度与可靠度进行区分提取,基于预处理数据分析了假冒绿通车在进出收费站时的时空特征、通行特征及其他特征。接着采用Borderline-SMOTE过采样法来平衡数据集,用ChiMerge算法来离散化连续型属性,并对离散的相关属性进行了关联项检验与共线性检验,以保证贡献程度大的属性与结果能够有效映射,提升属性与结果之间的关联程度与准确度。最后将通过关联项和共线性检验的自变量选入假冒绿通车逃费行为预测模型,并通过决策树对车辆类型分类来分辨假冒绿通车。利用历史绿通车通行数据集对比了逃费行为预测模型与其它模型的分类结果。结果表明:所提出的逃费行为预测模型准确率为83.4%,高于Logistic回归模型(61.8%)和随机森林模型(81%)。本研究模型能够有效预警假冒绿通车,在简化绿通车检查流程的基础上,降低假冒绿通车成功逃费情况的发生概率,具有实际应用意义。At present,the"green passage"policy is fully implemented.However,the detection is relatively lagging,and truck drivers are susceptible to counterfeit the toll-free logistics vehicles(TFLVs),causing huge losses to the operation and management agent.In order to improve the accuracy and efficiency of TFLVs inspection,a toll evasion prediction model for counterfeit TFLVs is established based on the historical TFLVs traffic dataset derived from the toll collection system of Sichuan expressway network.First,according to the importance and reliability,the data attributes are differentiated and extracted by using the data mining technology.Based on the preprocessed data,the spatiotemporal characteristics,traffic characteristics and other characteristics of the fake TFLVs are analyzed entering and exiting toll booths.Then,the data set is balanced by Borderline-SMOTE oversampling method,the continuous attributes are discretized by ChiMerge algorithm.In order to ensure the effective matching and improve the correlation between the large contribution attributes and the result,the correlation item test and the collinearity test are conducted on discrete related attributes.Finally,the independent variables that passed the correlation item test and collinearity test are selected into the toll evasion prediction model for counterfeit TFLVs,and the vehicle types are classified by decision tree to distinguish the counterfeit TFLVs.The classification results of the proposed prediction model and other models are compared by using the historical TFLVs traffic dataset.The result shows that the accuracy of the proposed prediction model is 83.4%,which is higher than that of Logistic regression model(61.8%)and random forest model(81%)respectively.The proposed model can effectively warn of fake TFLVs and reduce the occurrence probability of fake TFLVs evasion on the basis of simplifying the inspection process of TFLVs,which has practical application significance.

关 键 词：智能交通 假冒绿通车逃费行为 数据挖掘 历史绿通车通行数据 决策树 

分 类 号：U491.4[交通运输工程—交通运输规划与管理]