基于实测微观驾驶状态的交通安全风险分析及模型校正

发布时间：2018-11-22 15:39

【摘要】：随着经济的发展,越来越多的车辆进入普通家庭中。伴随着车辆的普及,交通安全问题日益严重。不安全的交通环境不仅会带来财产的损失,还会威胁人的生命安全。而在交通安全问题中,与事故发生后再补救相比,如何在事故发生之前及时采取措施,避免危险的发生则显得尤为重要。论文首先在NGSIM数据的基础上,分别对宏观与微观的交通状态和风险性进行了分析。而后根据Fuzzy C-means clustering method(简称FCM)聚类方法将实测数据划分了不同的聚类,并应用Helly模型对不同聚类进行了校正,并分析了不同风险形成的不同特点。最后将风险指标引入了 Helly模型中来避免出现较高的交通安全风险。仿真分析发现:引入风险指标之后,驾驶员能够根据自身驾驶状态及时调整驾驶行为来模拟避免潜在风险。其次,从NGSIM数据提取出了多个车辆组,分析了车辆组中头车的驾驶状态和风险对后续跟随车的影响,以及车辆组中相邻两辆前后车之间的相互影响。实测数据表明:头车的速度和速度差与跟随车的速度和车间距有明显的相关性,而且头车的风险大约能够影响到第5辆跟随车;因此,我们根据头车的速度和速度差,跟随车的速度和车间距进一步对车辆组进行聚类的划分。结果发现,随着跟随距离的逐渐增大,分类1的风险变化不大,分类2的风险逐渐减小,分类3的风险逐渐增加。整体来看,分类2的风险最高。最后我们对不同车辆组进行了模型校正,并分析车辆组中不同位置车辆的风险性。结果发现,对于不同位置的车辆,应特别关注速度差和速度的变化,来降低或避免风险。最后,我们利用中国合肥郊区跟驰实验数据分析了宏观和微观状态及风险的差别,发现微观状态能够更好地体现驾驶状态变化的瞬时性。聚类结果发现:分类3具有较小的速度差和较大的车间距,为低风险状态;分类1的速度差和间距都较小,而分类2的速度差和间距都较大,因此与分类3相比,都具有一定程度的风险。接着,我们将NGSIM数据和跟驰实验数据进行对比,以分析不同数据源的交通状态和风险差异。交通状态方面,宏观和微观状态下,NGSIM数据的速度、密度分布范围都比跟驰实验数据的小得多,而风险则比跟驰实验的略小。车辆组状态和风险相关性方面,跟驰实验数据头车速度与跟随车具有更强的相关性,而车辆组之间的风险相关性更小。在不同的分类下,NGSIM数据的间距分布范围更大,跟驰实验数据的速度差分布范围更大,因此相应的NGSIM数据的风险更低。
[Abstract]:With the development of economy, more and more vehicles enter ordinary families. With the popularity of vehicles, traffic safety problems are becoming more and more serious. Unsafe traffic environment will not only bring loss of property, but also threaten the safety of human life. In the traffic safety problem, how to take measures to avoid the danger is more important than remedying after the accident. Firstly, based on the NGSIM data, the traffic state and risk are analyzed. Then according to the Fuzzy C-means clustering method (FCM) clustering method, the measured data are divided into different clusters, and the Helly model is used to correct the different clustering, and the different characteristics of the formation of different risks are analyzed. Finally, the risk index is introduced into the Helly model to avoid the high traffic safety risk. The simulation results show that the driver can adjust his driving behavior according to his driving state to avoid the potential risk by introducing the risk index. Secondly, several vehicle groups are extracted from the NGSIM data, and the influence of the driving state and risk of the first vehicle in the vehicle group on the follower vehicle is analyzed, as well as the interaction between the two adjacent front and rear vehicles in the vehicle group. The measured data show that the velocity and velocity difference of the head car have obvious correlation with the speed and the distance of the vehicle, and the risk of the first car can affect the fifth car. Therefore, according to the speed and speed difference of the vehicle, the speed and the distance of the vehicle are further divided into clusters. The results show that with the increasing of the following distance, the risk of classification 1 does not change much, the risk of category 2 decreases gradually, and the risk of category 3 increases gradually. Overall, Category 2 has the highest risk. Finally, we calibrate the models of different vehicle groups and analyze the risk of vehicles in different positions. It is found that for vehicles in different positions, special attention should be paid to the variation of speed and speed to reduce or avoid risks. Finally, we analyze the difference between macro and micro states and risks by using the experimental data from the suburb of Hefei, China, and find that the microscopic state can better reflect the instantaneous change of driving state. The clustering results show that classification 3 has a small speed difference and a large vehicle spacing, which is a low risk state; The velocity difference and spacing of classification 1 are small, but the velocity difference and spacing of classification 2 are large. Therefore, compared with classification 3, both have a certain degree of risk. Then, we compare the NGSIM data with the experimental data to analyze the traffic state and risk difference of different data sources. In terms of traffic state, the velocity and density distribution range of NGSIM data is much smaller than that of car-following experiment data, and the risk is slightly smaller than that of car-following experiment. In the aspect of vehicle group status and risk correlation, the first car speed has a stronger correlation with the following vehicle, but the risk correlation between the vehicle group is less. Under different classification, the range of distance distribution of NGSIM data is larger, and the range of velocity difference distribution of NGSIM data is larger, so the risk of corresponding NGSIM data is lower.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U491

【参考文献】