【作者】 李敏；

【导师】 胡仕成；

【作者基本信息】 哈尔滨工业大学 ， 工商管理， 2021， 硕士

【摘要】 随着交通环境复杂性的增加,全国交通事故发生数、伤亡数和因此造成的经济损失同步攀升。同时,车联网技术的发展使得“人-车-环境”三者互联成为了一个交通系统,通过利用车联网技术搜集的海量数据,学者们得以采用数据挖掘技术对驾驶风险管理问题进行系统性的研究。但是大多数研究在构建驾驶风险预测模型时未考虑模型指标间的相关性、冗余性和特征对于预测驾驶风险的重要性程度,对驾驶风险进行研究时没有将驾驶员历史数据与当前实时驾驶情况综合起来考虑,缺乏适用于车联网环境的集成驾驶风险评估模型。为此,本文设计了一个基于特征选择的驾驶风险影响因素指标体系,并提出了一个对驾驶风险全过程进行管理的驾驶风险精准分级与实时预警集成模型。本研究设计的集成模型通过事前预先分级、事中实时监控并预警、事后更新历史数据的方式对驾驶风险全过程进行科学管理,而对驾驶风险进行分级是为了更好地管理风险,因此,必须为不同的风险情况设定不同的预警等级,这种差异化的预警策略有利于对驾驶员进行针对性的驾驶建议,最大程度降低驾驶风险。本模型旨在能够结合驾驶员历史数据和实时驾驶状态对驾驶员作出快速且精准的分级,并给出与等级相对应的风险预警,达到降低驾驶风险的目的。此外,模型还可以作为车险公司制定车险方案的依据,也可以为汽车生产厂商给客户提供个性化服务提供依据。实验结果表明,基于特征选择的驾驶风险影响因素指标体系对于集成模型的运行效率有提升效果。本文还对集成模型分类效果和风险预警效果进行了检验,验证了模型能够较好地根据驾驶员历史数据和实时状态对驾驶员做出精准分级和实时预警;验证了在接收到风险预警后,无论驾驶员是否有对行为做出调整、做出何种程度的调整,集成模型都可以准确识别出驾驶员实时状态的变化,并给出合理的风险预警等级和驾驶建议。更多还原

【Abstract】 With the increase of the complexity of the traffic environment,the number of traffic accidents,the number of casualties and the resulting economic losses rise synchronously.At the same time,the development of vehicle networking technology makes the "human-vehicle-environment" interconnection into a transportation system,by using the massive data collected by vehicle networking technology,scholars can use data mining technology to systematically study the problem of driving risk management.However,when constructing driving risk prediction models,most studies did not consider the importance of the correlation,redundancy and characteristics of model indicators in predicting driving risks.In the study of driving risk,the historical data of drivers and the current real-time driving situation are not considered,and there is a lack of integrated driving risk assessment model suitable for vehicle network environment.For this reason,this thesis designs an index system of influencing factors of driving risk based on feature selection and puts forward an integrated model of accurate classification and real-time early warning of driving risk to manage the whole process of driving risk.The integrated model designed in this study scientifically manages the whole process of driving risk by grading in advance,realtime monitoring and early warning in the event,and updating historical data afterwards,while the classification of driving risk is to better manage the risk.Therefore,different early warning levels must be set for different risk situations,and this differentiated early warning strategy is conducive to targeted driving advice to drivers.Minimize driving risk.The purpose of this model is to make a fast and accurate classification of drivers with the combination of driver historical data and real-time driving status,and to give the risk warning corresponding to the level,so as to achieve the purpose of reducing driving risk.In addition,the model can also be used as a basis for car insurance companies to formulate car insurance plans and can also provide a basis for automobile manufacturers to provide personalized services to customers.The experimental results show that the driving risk influence factor index system based on feature selection can improve the operation efficiency of the integrated model.This thesis also tests the classification effect and risk early warning effect of the integrated model and verifies that the model can make accurate classification and real-time early warning for drivers according to the driver’s historical data and real-time status.It is verified that after receiving the risk early warning,the integrated model can accurately identify the changes of the real-time state of the driver,regardless of whether the driver has adjusted his behavior or not.And give a reasonable risk early warning level and driving suggestions.更多还原