| The expanding scale of urban road network,increasing complexity of urban road network structure and increasing number of motor vehicles have brought a great challenge to the space-time balance of urban traffic flow,traffic safety and traffic environmental management.Route choice and guidance,which is a key factor in traffic assignment,playing an important role in alleviating urban traffic problems.With the popularization of intelligent transportation,various route guidance systems were proposed to provide recommended paths for travelers in city or highway.However,these route guidance systems adopted traditional path planning algorithm to generate the shortest path with the shortest time or distance.There is a big difference between travelers’ route choice in actual travel and recommended shortest path.Thus,route choice set should be based on a large number of travelers’ actual behaviors.Compare with cruising,taxi passengers carry operation is a trip with a certain purpose,this operation state could represent private cars’ trip in some extent.With the popularization of intelligent transportation,vehicle-borne GPS system has been widely integrated into taxi,which provide technical support for large-scale and large-scale acquisition of real route choice behaviors.Besides,most of taxi drivers are familiar with the road network and have rich driving experience,this could help them choose the most efficient path.Thus,this paper analyzes the route choice behavior of taxi passengers carry condition,which could also provide reference for the route choice guiding in urban roads.The main contents of this study are as follows:(1)Taxi GPS data of Xi’an was selected as the data foundation of this research.Considering the purpose of research,original GPS data set of taxi was preprocessed,and road network model of Xi’an was established based on openstreetmap(OSM)road network.Then the coordinate system of GPS data and road network model was unified by Python,and a flow model was put forwardt to match the GPS data with road network model by feature manipulate mngine technology(FME).(2)First,a difference-state method was put forward to identify the drop point and off point of taxi,the temporal and spatial distribution characteristics of pick-up point and drop-off point of taxi were analyzed.Then,the taxi pick-up and drop-off hotspots was mined by the method of DBSCAN spatial clustering algorithm,twelve pick-up hotspots and twelve drop-off hotspots were obtained.Finally,eleven drop-off hotspots of taxi were selected as the basis of hotspot OD matching and path information acquisition.(3)First,a hotspot OD matching algorithm was proposed to obtain a large number of paths between the same origin-destination,this algorithm contain two steps: 1)reverse recurrence of taxi drop-off points;2)re-clustering of drop-off points.Then,by the method of hotspot ODs matching algorithm,eighteen hotspot ODs were obtained based on eleven drop-off hotspots.Finally,travel information of each path between hotspot ODs was obtained by Python and road information of each path between hotspot ODs was obtained by FME.(4)According to the distribution of euclidean distance and volume,five schemes of taxi passenger-carrying route classification,which based on the circuity K-means clustering,were proposed.Then,the AIC principle was adopted to identified the best scheme among five taxi passenger-carrying route classification schemes.According to the best scheme,taxi passenger-carrying routes were divided into three categories: short distance,medium distance and long distance.Among eighteen hotspot ODs,three hotspot ODs were selected to analyzed the characteristics of the three types of taxi passenger-carrying routes.Then,based on the traditional multinominal logit(MNL)model,three PS-Logit models of taxi passenger-carrying route choice were proposed.Finally,from the aspects of route selection probability,PS-Logit and MNL model calibration results,marginal substitution rate(MSR),passenger-carrying route choice model of taxi was analyzed.(5)In order to verify the validity of three PS-Logit models,two verification method was adopted: 1)Qualitative analysis of model coefficients;2)Trial calculation of PS-Logit models combined with the hotspot ODs. |
