Urban Traffic Sensing And Taxi Demand Forecast Based On Taxi GPS Data

With the wide adoption of wireless communication networks and OBUs(Onboard Units),convenient and accurate traffic information service via VANET(Vehicular Ad-hoc Networks)has come true.Online navigation and route planning by traffic data mining are important applications,and has received wide attention from government,academia and industry.In our research,GPS equipped on vehicles is regarded as sensor node.Based on GPS data,we can actualize traffic sensing and taxi demand forecast,which is significant for route planning.Our main contribution is listed as follows:1.Traffic situation information can help the drivers make better route planning.For traffic management department,adoption of traffic sensing data can improve the traffic volume,and reduce traffic jam.However,both traffic sensing by vehicles and by road monitoring infrastructures face the following problem,i.e.data missing on the dimensions of time and space.It makes data reconstruction become a challenge for full-time and wide area traffic sensing.In this thesis,we try to solve the data reconstruction based traffic sensing under the framework of VANET.Our main research contributions are as follows:Mining the correlation of traffic data on consecutive days and doing principle component analysis of the traffic data,we reveal the hidden structure of the traffic data and propose a tensor-completion-based approach to estimate the vacant traffic information.Compared with the traditional matrix completion approach,it can exploit data pattern on more dimensions because tensor have more modes while matrix only has two modes.The tensor completion algorithm based on ADMM,i.e.HaTTC,outperforms FPCA in speed and accuracy in the simulation in the case of highway.Taking temporal continuity and data periodicity into consideration,we improve the accuracy.By adopting the half thresholding regularization,the algorithm gets accelerated by a great deal.We establish a VANET-based traffic monitoring system consisting of thousands of GPS-equipped vehicles.Based on the data,we achieved full-time traffic sensing by performing our algorithm.With its assistance,vehicles can conduct the route planning in a travel.2.Real-time taxi demand information can help the taxi drivers make reasonable route planning and increase their income.At the same time,passengers can save more time in waiting for taxi.However,taxi demand forecast are faced with many difficulties,such as uncertainty from discrete GPS data,high dynamics of taxi demand,complexity of traffic situation.Under the framework of VANET,our contribution for taxi demand forecast is as follows:We establish a new model-Taxi Demand Model(TD Model)to analyze and forcast taxi demand.Compared with other models,our model can present the components of the taxi demand in a more reasonable way.Besides,we identify taxi demand with a combined offline big data analysis and a real-time data stream,considering high dynamics of taxi demand.In TD model,two points should be highlighted.We present a novel parameter to quantify the difficulty in meeting a passenger,i.e.,how soon a taxi can pick up a passenger after entering a road segment at a time slot.With the adoption of the new parameter,the forcast accuracy is improved.Besides,we introduce geomegtric probability model to infer the passenger arriving moments.We establish a VANET-based taxi demand forcast system consisting of thousands of GPS-equipped vehicles.Based on GPS data,we testify our algorithm.