The Inference Of Public Transit Trip Purpose Baesd On The Data Fusion

With the rapid change of traffic demands,a large number of passive traffic data has been generated.The passive data has the characteristics of large quantity,continuation and accuracy compared with the traditional data of resident travel survey.Developing a new traffic survey method with the passive data is an inevitable choice in the age of big data.However,the inherent weakness of passive data is that it does not contain the attribute of trip purpose.Therefore,how to use a reasonable method to predict the purpose of trip and improve the trip information of the passive data is the core issues in such environment.This study uses the traditional resident survey data as the basic data source.It uses different machine learning methods,such as decision tree,support vector machine,neural network,Bayesian network and random forest.It builds different models with the different input variables so as to explore the practicability of the methods with the passive data only.This study tries to compare the performance of different methods of support vector machine,random forest,decision tree,neural network and Bayesian network to get the best model in trip purpose inference.Decision tree,random forest and support vector machine are the top three methods in all.On the basis of such conclusion,the study uses the intelligent traffic card of Chengdu with 1 day to predict the purpose of each trip.Finally,the study compares the predicted travel ratio with the residents' trip survey results of Chengdu to validate the model.