【作者】 王瑞峰；

【导师】 黄妙华；

【作者基本信息】 武汉理工大学 ， 车辆工程， 2023， 硕士

【摘要】 驾驶行为决定了无人驾驶汽车的安全性、乘客舒适性等性能,对驾驶行为进行科学的评价是很有必要的。针对目前驾驶行为评价维度较为单一、评价指标太少且不够系统、未考虑行驶工况的问题,本文结合国家重点研发计划课题“典型场景下纯电动汽车推广及商业模式关键影响因素互动分析研究”(编号:2018YFC0808405)和“电动汽车大规模分时租赁和集成示范运营车载系统研发及分析”(编号:2015BAG08B02)采集的实车行驶数据,开展行驶片段样本数据库建立、行驶工况划分方法研究、驾驶行为评价模型搭建三方面的研究,全文工作如下:提出起步-中途-停车三阶段的行驶片段划分方法,并建立行驶片段样本数据库。首先研究数据采集频率对研究驾驶行为的影响,确定在行驶片段时长超过70秒,且采集频率小于等于0.1HZ时,驾驶行为特征参数相较1HZ差距在5%以内,能够表征行驶片段的驾驶行为;然后根据车辆起步、中途和停车三个阶段对1000辆纯电动车一年的525814152条自然驾驶数据进行了行驶片段划分,得到1308621个行驶片段,设定了46个行驶特征参数,得到行驶片段样本数据库,为工况划分以及驾驶行为评价奠定了数据基础。提出基于DBSACN-Kmeans++聚类的行驶工况划分方法。首先构建了测试样本数据集,对比了常用聚类算法的聚类效果,结果表明,提出的DBSACN结合Kmeans++的聚类算法轮廓系数最接近1,CH分数最大,DBI指数最小,对高密度的样本数据集的适应性更好;然后对能够表征行驶工况特点的特征参数组合进行主成分分析降维和聚类;最后划分了拥堵低速、顺畅中低速和顺畅高速行驶工况三个工况,统计三种工况下的车辆允许的最大运行速度和车速标准差的最大限值,为驾驶行为评价提供对应的评价标准。提出基于改进TOPSIS法的驾驶行为综合评价方法。首先制定评价指标选取原则并建立了驾驶行为综合评价指标体系,确定驾驶行为安全性、乘客舒适性和经济性3个一级指标,8个二级指标,16个三级指标。然后对驾驶行为评价指标重要程度进行了问卷调查,利用层次分析法对一级指标和二级指标进行了主观赋权,提出结合指标的熵值、波动性和相关性的改进CRITIC法,并结合层次分析法对三级指标混合赋权。最后,在传统逼近理想解TOPSIS法上,提出考虑向量余弦值的改进TOPSIS法作为驾驶行为综合评价方法。对自然驾驶和无人驾驶实车驾驶行为进行评价,得到驾驶行为综合得分,无人驾驶实车评价结果与乘客主观评价相同,验证了评价模型的科学性和有效性。更多还原

【Abstract】 Driving behavior determines the safety and passenger comfort of driverless vehicles,and it is necessary to conduct a scientific evaluation of driving behavior.Aiming at the current problems of relatively single dimensions of driving behavior evaluation,too few and insufficient evaluation indicators,and not considering driving conditions,this article combines the real vehicle driving data collected by the national key research and development plan topic "Interactive Analysis and Research on Key Influencing Factors of Pure Electric Vehicle Promotion and Business Model in Typical Scenarios"(Project No.: 2018YFC0808405)and "Research and Analysis of Onboard System for Large Scale Time-sharing Lease and Integrated Demonstration Operation of Electric Vehicles"(Project No.:2015BAG08B02)to conduct the establishment of a driving segment sample database,research on driving cycle division methods The research on three aspects of building a driving behavior evaluation model is as follows::The method of driving segment division in three stages of starting,midway and stopping is proposed,and the sample database of driving segment is established.First,study the impact of data acquisition frequency on driving behavior,and determine that when the driving segment duration is more than 70 s,and the acquisition frequency is less than or equal to 0.1HZ,the difference between the driving behavior characteristic parameters and 1HZ is within 5%,which can represent the driving behavior of the driving segment.Then,525814152 pieces of natural driving data were divided into driving segments based on the three stages of vehicle starting,midway,and stopping,resulting in 1308621 driving segments.46 driving characteristic parameters were set,and a sample database of driving segment characteristics was obtained,laying a data foundation for driving cycle division and driving performance evaluation.A driving cycle division method based on DBSACN-Kmeans++clustering is proposed.First,the test sample data set is constructed,and the clustering effects of common clustering algorithms were compared.As a result,it was found that the cross section coefficient of the combination of DBSCAN and kmeans + + was close to 1,and the DBI index is the smallest,which is better adaptive to the high-density sample data set.Then cluster the sample database of driving segments,and finally divide the three driving conditions of congested low speed,smooth medium low speed and smooth high speed,determine the maximum allowable operating speed of vehicles under different driving conditions and the maximum limit of standard deviation of speed,and provide relevant evaluation criteria for driving behavior evaluation.An integrated driving behavior evaluation method based on improvement of TOPSIS is proposed.Firstly,the principles for selecting evaluation indicators were formulated and a comprehensive evaluation index system for driving behavior was established,including three first level indicators for driving behavior safety,passenger comfort,and economy,eight second level indicators,and sixteen third level indicators.Then,a questionnaire survey was conducted on the importance of driving behavior evaluation indicators.The first and second level indicators were subjectively weighted using AHP,and an improved CRITIC method combining entropy,volatility,and correlation of the indicators was proposed.The third level indicators were mixed weighted using AHP.Finally,based on the traditional approximate ideal solution TOPSIS method,an improved TOPSIS method considering the vector cosine value is proposed as a comprehensive evaluation method for driving behavior.The comprehensive score of driving behavior was obtained by evaluating the natural driving and driverless real vehicle driving behaviors.The evaluation results of driverless real vehicle were the same as the subjective evaluation of passengers,verifying the scientificity and effectiveness of the evaluation model.更多还原