Research On Traffic Travel Mode Identification Based On GPS Trajectory Data

With the rapid development of economic and urbanization,traffic congestion,traffic accidents and traffic environment have become common urban diseases in China.Scientific traffic planning and management are considered as effective methods to solve these problems.While the residents travel information has become the support of scientific traffic planning and management.The most widely used method for the investigation of travel information is the OD survey.But it is influenced by consciousness of investigators.Omissions and errors will affect the quality of the survey data.Besides,it has high cost,heavy workload,low recovery rate and long cycle.Nowadays,the expansion of global positioning system(GPS)and personal travel data survey based on GPS make it possible to get more accurate and complete personal travel behavior information.It has been a new effective way to gain traffic information because of its high efficiency and high precision.It is an important part for personal travel information to obtain more accurate and complete information of travel behavior though mining GPS data.Especially,the transform point recognition has become an important part of transportation way identify and research topic.This paper proposes two methods to identify transform points based on similarity measure and window,multi-section method and moving window method.Then these methods are compared and better method is applied to identify the traffic way to get travel mode.Then this paper extracted characteristic parameters of travel mode and identify traffic mode by BP neural network,decision tree,KNN and SVM.The proposed algorithm is verified by GPS trajectory data of Geolife engineering.After comparison,multi-section method turns out better than moving window method.F-score is close to 80%and its recall rate is close to 90%.In recognition of travel mode,SVM is the best way and its training set and test set are 92.75%and 88.77%.At last,the recognition accuracy increases from 88.77%to 91.84%based on time index to evaluate the recognition results.So the proposed method has high recognition accuracy.In this paper,the effectiveness of traffic mode recognition algorithm based on GPS data is verified by experiments.The research conclusion can be applied to analyze the characteristics of traffic information.They will be important data support for traffic planning and management.