Predicting Trip Destination With Markov Model

In recent years,GPS data has become a new way of the analysis of resident travel behavior with the widespread use of mobile phone software and GPS(global positioning system)navigation devices.Travel information is extracted with the GPS data mining,and is used to predict trip destination.It is the more complex process in the studies of travel behavior forecasting.Through predicting the trip destination,the resident spatial distribution characteristics can be analyzed accurately.Then urban transportation system can be planned more reasonable.It is the effective way to improve traffic problem,at the same time has far-reaching significance to ensure the sustainable development of urban traffic.This study presents the trip destination prediction system with Markov model through studying the analysis of GPS data processing.Firstly,the study investigates and processes the trip GPS data.It includes translating GPS points into continuous trips and determining the destinations of trips.Then the study divides the trips into weekdays' trips and weekends' trips,and extracts the frequently-visited destinations and support points with reasonable parameters.The modeling data of weekdays and weekends can be obtained,respectively.Secondly,the study develops pre-trip destination prediction based on Markov chain model and during-trip destination prediction based on Hidden Markov model.A pre-trip destination prediction model is constructed and calibrated for trips in weekdays and weekends,respectively.And a during-trip destination prediction model is also constructed and calibrated for trips in weekdays and weekends,respectively.Then the example analysis can be implemented with resident trip GPS data.The results show that the prediction precision of pre-trip and during-trip are high enough and more preferable than existing models due to separately modeling of weekdays' and weekends' trips.In comparison,the during-trip destination prediction has better prediction effect because of the use of support points.So trip destination prediction methods proposed in this paper can be applied to predict resident trip destination,either in anticipation of trip destination before a trip or the real-time trip destinations correction during a trip.The Markov model about trip destination prediction proposed in this paper does not need the traditional household trip survey and road map,reduces the complexity of data investigation and simplifies the process of data processing.Research results can be applied to vehicle navigation equipment or mobile navigation software for the real-time trip destination prediction of pre-trip and during-trip.This will save the step of inputting trip destination for increasing the speed of navigation operation and the driving safety.And the trip route,traffic condition and some facilities around the destination can be recommended to travelers for satisfying the travel demand.At the same time,the trip destination prediction presents a wide application in transportation planning and management,such as resident spatial distribution analysis,travel demand forecast,crowded location forecast analysis,etc.Moreover,the study on the disaggregate behavior model of resident trip destination has a certain significance for prompting traffic demand forecast transform from traditional four-stage model to advanced disaggregate model.