基于变量选择和核极限学习机的交通事件检测  

Traffic incident detection based on variable selection and kernel extreme learning machine

作　　者：商强[1] 林赐云[1,2] 杨兆升[1,2] 邴其春[1,3] 邢茹茹 

机构地区：[1]吉林大学交通学院,吉林长春130022 [2]吉林大学吉林省道路交通重点实验室,吉林长春130022 [3]青岛理工大学汽车与交通学院,山东青岛266520

出　　处：《浙江大学学报（工学版）》2017年第7期1339-1346,1445,共9页Journal of Zhejiang University：Engineering Science

基　　金：国家"十二五"科技支撑计划资助项目(2014BAG03B03);国家自然科学基金资助项目(51408257;51308248)

摘　　要：为了进一步提高交通事件检测的效果,提出基于变量选择和核极限学习机(KELM)的自动事件检测(AID)算法.根据交通事件上、下游交通流参数的变化特点,构建较全面的交通事件检测初始变量集.采用随机森林—递归特征消除(RF-RFE)算法,从中选择重要变量.以重要变量作为输入,训练KELM并通过万有引力搜索算法(GSA)优化参数.使用美国I-880数据库,对AID算法的效果进行验证和对比分析.因为数据库中的事件样本数远少于非事件样本数,采用SMOTE平衡两类样本.结果表明,使用重要变量能够提高交通事件的检测效果,KELM的检测效果优于反向传播神经网络(BPNN)和支持向量机(SVM).An automatic incident detection（AID）algorithm was proposed based on variable selection and kernel extreme learning machine（KELM）in order to further improve the performance of traffic incident detection.A relatively comprehensive initial variable set was constructed for traffic incident detection according to the changing characteristics of upstream and downstream traffic parameters in a traffic incident,and important variables were selected from the initial variable set by the random forest-recursive feature elimination（RF-RFE）algorithm.Then the important variables were used as input of KELM and the KELM was trained.The parameters were optimized by the gravitational search algorithm（GSA）.The I-880 database of the United States was used to validate and comparatively analyze the performance of the proposed AID algorithm.Because the number of incident samples is much less than the number of incidentfree samples in the database,the SMOTE is used to balance the two kinds of samples.Results show that using important variables can improve the performance of traffic incident detection,and the performance of KELM is better than that of the back propagation neural network（BPNN）and support vector machine（SVM）.

关 键 词：交通工程 交通事件检测 变量选择 随机森林 极限学习机(ELM) 

分 类 号：U491[交通运输工程—交通运输规划与管理]