摘要
交通事故风险与交通流状态存在显著关系,危险的交通流状态易诱发交通事故。为降低交通事故的发生率,保障交通系统的运营安全,通过实时监测道路交通流参数的变化情况,构建支持向量机模型(Support Vector Machine,SVM)对易引发交通事故的危险交通流状态进行识别,及时预判潜在的交通事故。首先选用实际交通事故发生前的交通流状态作为危险交通流状态的判别标准,分析交通流特性,提取24个交通事故前兆特征变量。为提高模型性能,降低其计算复杂度,设计相关性选择算法(Relevance Selection Algorithm,RSA)对24个特征变量进行降维,该算法充分考虑各前兆特征变量与交通流状态类别的相关性以及各前兆特征变量之间的相关性,最终保留4个交通事故前兆特征变量。接着采用改进的网格搜索算法优化支持向量机模型的惩罚参数C和核函数参数γ,参数寻优效率比传统的网格搜索算法提高了98.3%,极大地节省了搜索时间。最后根据所构建的危险交通流状态实时识别模型,以某城市快速路的事故数据为例进行数值计算。结果表明:该模型具有较快的危险交通流状态识别能力和潜在交通事故的预警能力,且识别正确率比经典的K近邻算法提高5%、比BP神经网络算法提高22.3%。该方法能有效地对危险交通流状态进行实时识别,可为交通管理部门制订城市快速路交通事故风险管控方案提供理论依据。
There is a significant relationship between traffic accident risk and traffic flow state.Dangerous traffic flow state is easy to induce traffic accidents.In order to reduce the incidence of traffic accidents and ensure the safety of traffic system operation,the dangerous traffic flow states which are prone to traffic accidents are recognized by the constructed SVM and real-time monitoring the changes of road traffic flow parameters to timely predict the potential traffic accidents.First,selecting the traffic flow state before the actual traffic accident as the criterion of the dangerous traffic flow state,the traffic flow characteristics are analyzed,and 24 traffic accident precursor feature variables are extracted.In order to improve the performance of the model and reduce its computational complexity,a RSA is designed to reduce the dimension of the 24 characteristic variables.The algorithm fully considers the correlation between each precursor characteristic variable and traffic flow state category as well as the correlation between different precursor characteristic variables,and finally retains 4 traffic accident precursor characteristic variables.Then,the penalty parameter C and the kernel function parameterγof the SVM are optimized by using the improved grid search algorithm.The efficiency of parameter optimization is increased by 98.3%than that of the traditional grid search algorithm,which greatly saves the search time.Finally,according to the constructed real-time recognition model of dangerous traffic flow,taking the accident data of an urban expressway for example,the numerical calculation is conducted.The result shows that(1)the model has the ability of rapid recognition of dangerous traffic flow status and early warning ability of potential accidents,the recognition accuracy rate is 5%higher than that of the classic K nearest neighbor algorithm,and 22.3%higher than that of the BP neural network algorithm;(2)the method can effectively recognize the real-time dangerous traffic flow state,and can provi
作者
孙然然
张静萱
朱广宇
SUN Ran-ran;ZHANG Jing-xuan;ZHU Guang-yu(Beijing Research Center of Urban Traffic Information Intelligent Sensing and Service Technologies,Beijing Jiaotong University,Beijing 100044,China;Key Laboratory of Big Data Application Technologies for Comprehensive Transport of Transport Industry,Beijing Jiaotong University,Beijing 100044,China;Planning and Standard Research Institute of National Railway Administration,Beijing 100055,China)
出处
《公路交通科技》
CAS
CSCD
北大核心
2021年第10期120-128,共9页
Journal of Highway and Transportation Research and Development
基金
国家自然科学基金项目(61872037,61833002_3)
中央高校基本科研业务费专项资金项目(2021JBM403)。
关键词
交通工程
实时识别
支持向量机
相关性选择算法
危险交通流状态
交通事故
traffic engineering
real-time recognition
support vector machine(SVM)
relevance selection algorithm(RSA)
dangerous traffic state
traffic accident
