Clustering-based Travel Pattern For Individual Travel Prediction Of Passengers By Using Transit Smart Card

Individual travel prediction is very important for the construction of intelligent urban transportation system.Most of the previous researches in this field used mobile signaling data to predict the user’s next location.With the development of intelligent transportation system,smart cards(IC cards)provide massive bus passenger travel data based on travel decisionmaking for the study of individual passenger travel.On this basis,we first predict whether the passenger will travel,and then,if so,predict the probability distribution(travel time,travel origin,travel destination)of the next trip conditional on the previous one.In the past,the research methods for individual travel prediction mainly used Markov chains,and many scholars focus on the improvement of algorithms,but few researches pay attention on the mining of data information.In this paper,the concept of travel pattern is introduced into the field of individual travel prediction of frequent bus passengers.The travel pattern of passengers refers to the trip with similar boarding time and similar boarding and alighting stations of the same person.Through clustering the travel pattern by DBSCAN algorithm,the regularity of passenger travel can be better exploited and travel information can be integrated into a unified unit as well.In the process of predicting the individual trips of bus passengers,this paper first uses a binary prediction model based on historical travel probability to predict whether passengers will travel.Then the N-Gram algorithm is used to predict the passenger’s next travel pattern(or travel attributes).This paper uses the IC card data of frequent bus passengers in Chengdu in May 2019 to compare the methods based on travel chain and travel pattern from three levels:attributes prediction,boarding station and boarding time joint prediction,and full travel attribute joint prediction.Through the comparison of the prediction results,it can be found that the methods based on the travel pattern can effectively improve the prediction accuracy,especially the prediction results of the boarding time attribute can be greatly improved.On this basis,this paper further proposes active area-based method and anchor points-based method.Through the relaxation of the spatial threshold and the further screening of regular travel,these two methods can not only increase the proportion of travel pattern estimation,but also effectively improve the prediction accuracy of travel attributes in regular travel.