Urban Road Transportation Analysis Based On Floating Car Data

Intelligent transportation system(ITS)is considered as one of the effective methods of relieve the urban traffic problems such as traffic congestion,accidents and air pollution caused by motor vehicle exhaust.By gathering and analysising the real-time data of urban road network,short-time traffic flow forcasting could estimate the state of traffic flow for a few minutes in future and provide support to intelligent transportation control.As an important transportation research field,urban transportation analysis can provide effective support for intelligent traffic guidance and user route choice.Compared with other methods,support vector machines have more advantages to deal with the problems of under-fitting,local optimal and small samples.It can achieve pretty good performance in short-term traffic flow prediction.This thesis addresses the problem of urban transportation analysis and short-term traffic flow prediction from GPS data.A support vector machine regression model is built to predict the short-term flow of any local road of transportation network by mapping the GPS data information into whole road network.The PageRank algorithm is applied to deal with the sparse data,data incomplete and noise problems.With the complete weight annotation of the whole road network,optimal routes can be efficiently computed by routing algorithms and other transportation analyses can be carried out precisely.The main work is as follows:(1)Data preprocessing such as cleaning and normalization methods are used to suppress noise and emphasize the information of raw Xi’an taxi GPS data.The thesis also investigates the question of how much floating car size are needed to provide enough traffic speed information.Empirical study showed that the current GPS samples are suitable for urban traffic analysis.(2)A support vector machine regression model is developed to achieve the traffic states identification.The model use temporal,spatial,historical correlation and their combination for the input parameters.Best prediction performance can be achieved by using temporal-spatial information.(3)A modified weighted PageRank algorithm is applied to solve the problems of uniform,missing and the inaccurate problem of GPS data.By smoothing the incomplete and noise GPS data,the performance of traffic flow prediction is improved greatly.The prediction does not only pay attention to the local traffic flow information,but also consider the conditions of whole traffic network.An example of routing choice offers evidence that the model is effective.