Traffic Prediction And Path Selection For Urban Road Networks Based On The Data Collected By Mobile Phones

ABSTRACT:Intelligent Transportation Systems (ITS) is nowadays accepted internationally as the best way to solve traffic jams, decrease traffic accidents and lessen the emission and pollution, the premise and main point of which is road traffic informations. It is an important branch of ITS to get accurate traffic data within a short time and swiftly and exactly turn it into the optimal paths of drivers.With the increasing popularity of smart phones these years, facilities in communication as well as large amounts of traffic data produced in phones are brought to us. Making short-term forecasts and route guidance in road network system by collecting and handling data in cellphones becomes a cutting-edge topic in ITS. This paper mainly studies the modeling method and algorithm in making forecasts and route guidance in road network by utilizing traffic data in smart phones. There are three main contends discussed.1. Considering the features of road network and traffic data in cellphones and introducing the idea of virtual trip lines (VTLs), the paper designs a Spatio-Temporal Auto Regressive Moving Average (STARMA) model for forecasting travel time and a Logistic Regression (LR) algorithm for predicting congestion degree in road network.2. A path selection algorithm based on connection reliability is put forward using STA RMA and LR. It takes the probability of congestion from LR as the weight of travel time of the induced path, making the path with less or no traffic jams more inclined to be chose when selecting the optimal path and becoming a solution to the problem of sharp increase of local vehicles due to blindly pursuing the minimum time cost, which eventually affects the connection reliability of road network.3. The paper achieves a simulation on the process of collecting traffic data in cellphones by programming in Paramics software, in which a road network model is built and simulated. The results produced by the STARMA model and the LR algorithm have been compared with real data as well as the results produced by other combining algorithm.