Research On Urban Travel Time Prediction Methods With Floating Car Data

Intelligent Transportation System(ITS)is an efficient,safe,fast and comfortable transportation system that is based on the exsisting transportation insfrastructure.ITS tries to reduce the traffic accident,decrease the pollution index and improve the effciecy of the transportation system by using the technology of multi-media,wireless communication,automation and information processing.ITS can optimize the traffic flow and reduce the traffic jams with a highly efficient management system that utilizes the exsisting transportation infrasture in an effective way.Travel time prediction is a key technology in ITS and it's also the foundation of many other ITS applications.The precise predition results can also provide guidance for the traffic management and control systems.In this dissertation,a deep research is conducted to solve the problems in the travel time prediction of complicated urban traffic network,and the main research results are as follows:1.A digital map based on a two-line road model is built to solve the problem that the current digital map in traffic applicationds can't represent the characteristic of the urban road network.A two-line road model is constructed to meet the demands of the urban road network that some special roads(such as roundabout and bridges)are represented and some forbidding rules are expressed.In this model,one road at an intersection can be forbidden to pass without affecting other roads of this intersection.Moreover,the travelling rules of some special roads(such as roundabout and bridges)are well represented in this model.2.A map matching algorithm based on computing geometry is proposed in this paper to solve the time efficiency and robust problems of the exsisting algorithms.The methods to determine the relations between line segements and polygons in computing geometry are used to determine the candidate roads.When determing the matching road,a confidence zone is built for every candidate road and the historical information are also utilized.3.A data interpolation algorithm is proposed in this paper to solve the data missing problem in the floating car data.First,a data missing analysis is conducted with exsisting floating car data.Then,the traffic information is generated from the floating car data with the vehicle-tracking method and the concept of speed matrix is proposed to reflect the temporal-spatial relations between the traffic information.Finally,the hidden information and temporal-spatial relations between the traffic information are revealed with a PCA based algorithm.Moreover,the revealed information is utilized to interpolate the missing data.4.A travel time estimation algorithm based on particle filter is proposed to solve the problem of current estimation algorithms.First,a large number of particles are initialized to represent some traffic pattern in the historical data.Then,all the paticles in the above steps are used to model the traffic trend in the historical data with weight,while tradition methods use state-transition function to model the traffic trend.Moreover,the weights of every particle are updated when a measurement comes.The updated weight reflects the similarity between the current traffic pattern and the traffic pattern represented by the particle.Next,all the particles with low weight are resampled to solve the degeneracy problems and all the particle with high similarity with current traffic pattern are used for estimation.Finally,the weighted sum of all the particle is chosen as the estimation results.5.A deep learning based travel time prediction method is proposed to solve the problem raised by the complex environment of uban transportation system.First,a correlation analysis is conducted to determine the input road charcteristics and the length of input traffic data.Then,the stack autoencoder is utilized to predict the travel time with historical traffic data and road charcteristics.Finally,the deep belief network is used for the data-fusion of the above prediction results,which provides the final prediction result.