基于GPS数据的公交站运行状态分析  

Analysis of bus-stop operating state based on GPS data

作　　者：黄洪滔 肖梅[1] 刘倩[1] 明秀玲 边浩毅 HUANG Hongtao;XIAO Mei;LIU Qian;MING Xiuling;BIAN Haoyi(College of Transportation Engineering,Chang’an University,Xi’an 710064,Shaanxi Province,P.R.China;Zhejiang Institute of Mechanical&Electrical Engineering,Hangzhou 310053,Zhejiang Province,P.R.China)

机构地区：[1]长安大学运输工程学院,陕西西安710064 [2]浙江机电职业技术学院,浙江杭州310053

出　　处：《深圳大学学报（理工版）》2023年第3期326-334,共9页Journal of Shenzhen University(Science and Engineering)

基　　金：浙江省“尖兵”“领雁”研发攻关计划资助项目(2022C01105);陕西省社会科学基金资助项目(2022F021)。

摘　　要：为有效提高公交站点的运行效率,对公交站的运行状态进行识别、预测及影响因素分析,以中国西安市公交车全球定位系统轨迹数据为例,建立平均服务时间和服务车数特征参数反映公交站的运行状态,并通过分析站点内公交车辆速度、里程及加速度之间关系计算站台服务时间.使用Hopkins统计量和轮廓系数分析可聚性和聚类数,结合高斯混合模型(Gaussian mixture model,GMM)对公交站运行状态进行识别分类.构建SMOTEENN-XGBoost(synthetic minority oversampling technique edited nearest neighbours-extreme gradient boosting)站点运行状态预测模型,引入可解释机器学习框架SHAP(Shapley additive explanation)分析站台属性、道路及环境对模型的影响.结果表明,公交站运行状态可分为3类,类型Ⅰ的平均服务时间最长,类型Ⅱ的平均服务时间和服务车数最少,类型Ⅲ的服务车数最多;所建立SMOTEENN-XGBoost模型的准确率为94.68%,精确率为94.69%,召回率为91.04%,F1分数为92.26%,与极限梯度提升(extreme gradient boosting,XGBoost)、逻辑回归(logistic regression,LR)、随机森林(random forest,RF)、梯度提升决策树(gradient boosting decision tree,GBDT)和k近邻(k-nearest neighbors,KNN)5种模型对比,本模型能够精准预测站点运行状态;对站点运行状态具有影响作用的因素按照重要程度由大到小依次为线路数、有无公交专用道、泊位数、站台设置方法、站台几何形状、车道数、站台设置位置、是否工作日、时段及天气类型.研究结果可为公交站点设计优化提供一定参考依据.In order to effectively improve the operational efficiency of bus stops,this study conducts identification,prediction and analysis of the influencing factors of bus stops with different operational states.Taking the global positioning system(GPS)tracing data of bus vehicle in Xi’an as an example,the characteristic parameters of bus stop operating state including average service time and number of service vehicles were established,and the relationship between bus speed,mileage and acceleration within the bus stops was analyzed to calculate the service time.The Hopkins statistics and silhouette coefficient were used to analyze the clustability and determine the number of clustering categories,and three types of bus stop were identified based on Gaussian mixture model(GMM).The synthetic minority oversampling technique edited nearest neighbours-extreme gradient boosting(SMOTEENN-XGBoost)model was constructed to predict the bus stop operating state,and interpretable machine learning framework named Shapley additive explanation(SHAP)was employed to analyze the influence of three aspects:station attributes,road factors,and environmental factors on the operating state of bus stop.The results show that typeⅠhas the longest average service time,typeⅡhas the least average service time and number of service vehicles,and typeⅢhas the highest number of service vehicles.Compared with extreme gradient boosting(XGBoost),logistic regression(LR),random forest(RF),gradient boosting decision tree(GBDT)and k-nearest neighbors(KNN),the established SMOTEENN-XGBoost model can accurately predict the operating state of bus stop with an accuracy of 94.68%,precision of 94.69%,recall of 91.04%,and F1 score of 92.26%.The factors that influence the bus-stop operating state in descending order of importance are:the number of lines,the presence of bus lanes,the number of parking spaces,the method of platform installation,platform geometry,the number of lanes,the location of the platform,working day,the time of day and the type of weather.

关 键 词：城市交通 公交站点 运行状态 XGBoost模型 全球定位系统(GPS)数据 可解释机器学习 

分 类 号：U491.1[交通运输工程—交通运输规划与管理] TP391[交通运输工程—道路与铁道工程]