A Method Of Bus Travel Time Prediction Based On Gradient Boosting Regreesion Tree

In recent years,urban traffic problems,represented by traffic congestion,have generally become blocks for the development of a number of metropolises.In order to alleviate that,increasing the supply of urban transit and vigorously developing intelligent transportation system have become the main solution.And as an important indicator of ITS,bus travel time prediction has attracted more and more scholar's attention.At present,these studies mainly focus on the use of traditional prediction methods,such as Kalman filter model and support vector machine model based on a few basic features of the travel time prediction method.Compared with the traffic situation in the past,the current traffic situation has great differences in terms of data volume,non-linear and dynamic characteristics of public transport.Based on the given relevant data,the travel time of urban public transport on different bus routes is predicted,that is,the problem of travel time prediction of urban public transport studied in this paper.Accurate knowledge of the problem of bus travel time prediction can provide effective information for ITS,improve the scheduling optimization level of ITS,and alleviate traffic congestion.Aiming at the bus travel time prediction,this paper first analyses its influencing factors.According to the influencing factors,data preprocessing and feature analysis are carried out for the acquired data,and then different types of feature sets are constructed through data mining,such as basic feature set,historical statistical feature set,weather feature set,clustering feature set obtained by K-means and sliding window feature set obtained by sliding time window.And then two method of bus travel time prediction are designed based on gradient boosting regression tree model(GBRT)and light gradient boosting Machine model(LightGBM).Finally,taking GPS data and weather data of a city as experimental objects,the prediction model is constructed.The experimental results show the accuracy of LightGBM model is better than GBRT.But overall the accuracy of both methods are good.At the same time,LightGBM's training speed is very fast,which suits well the prediction of big data.Also,it can provide help for the command and dispatch of ITS.