Modeling Taxi Customer-Searching Behavior Using Frequent Sequence Mining And Two-Layer Logit Model

As a kind of urban public transport,taxi has more flexibility than other public transport modes.On the one hand,the flexibility of taxi complements other modes of public transport and provides"point-to-point"services to residents.On the other hand,however,problems caused by no-load taxis,such as road occupancy,traffic congestion and urban air pollution,also bring distress to transport operations.Therefore,studying the taxi cruising behavior and accurately modeling it is beneficial to the study of taxi operation management,scheduling and auxiliary systems,help to reduce the no-load rate,improve the operational efficiency of taxi drivers and alleviate the problems caused by no-load taxis to urban traffic.Based on the data of taxi GPS traces,this paper studies the modeling method of taxi searching behavior,which includes the following three aspects.1.A method for generating paths'choice set for taxi driver's customer searching modeling based on frequent sequence is proposed.PrefixSpan algorithm is used to mine the frequent sequence of taxi sections by utilizing the high similarity of searching trajectories between the same starting and ending points and the high difference of passing frequency of each road.The frequent road segment sequence is used as a"channel"for path planning,and the different paths in the planned paths are extracted.Finally,a set of candidate paths matching the real path is obtained.Experiments show that,according to the similarity index of the trajectory,the degree of similarity between the real trajectories and the candidate paths of most region pairs is greater than 0.6,and about 60%of the region pairs have a similarity greater than 0.7,which prove that the path sets generated by the method proposed in this paper is similar to the real trajectorys.2.By comparing and analyzing the modeling results of taxi customer-searching behavior using route selections'cost and benefit features respectively,it is found that the benefit features of the routes are important factors influencing drivers'route selection.Compared with the general path selection modeling,which only considers the cost features such as distance,travel time,turning times,this paper believes that drivers need to find passengers in the shorest time from the perspective of taxi customer-searching behavior theory,and adds the benefit characteristics of each path such as the number of passengers and the historical passenger travel distance et al.,which expands the path selection modeling features.Experiments show that after introducing the benefit characteristics,the R~2 mean value of the model is improved by0.053-0.066 which means that the fitting of the taxi customer-searching behavior by the model is more accurate.3.Methods based on two-layer logit model of simulating taxi searching behaviors are presented.Logit models of taxi searching behavior are constructed on regional scale and road scale respectively.In regional scale,MNL model is used to simulate the regional choice of taxi searching behavior.In road scale,PSL model is used to simulate the path choice.Experiments show that for the regional selection model,more than 80%of the regional models'R-square is greater than 0.6.For the path selection model,the region pairs with the models'R-square greater than 0.4 is 70.9%-79.9%,which is significantly higher than the comparison model.