Research Of Vehicle Mobility Patterns Based On Statisticle Probility

With the tremendous increase in the number of vehicles,the research of vehicle mobility is getting more and more important.The research on vehicle mobility will have far-reaching implications in many fields such as vehicular Ad-hoc network in wireless communication,urban traffic,Internet car-hailing and taxi operation,auto-driving filed etc.In order to bring benefits to these fields,this thesis conducts the research from the perspectives of feature analysis and mobility prediction respectively,aiming to extract effective information through the analysis of the real data.And a new model of vehicle trajectory prediction based on diffusion kernel is proposed.The main work and contributions are written as follow:(1)Vehicle mobility analysis based on real data:Based on the real taxi data of Beijing and the simulated base station data,this thesis analyzes the spatial distribution of the vehicles in time domain.The analysis can optimize the deployment solutions for roadside wireless communication units.Then this chapter analyzes the vehicle trajectory with K-means cluster method and then analyzes the passenger number and traffic jam distribution around Beijing.Finally gives out the direction of improving taxi income.(2)Research on vehicle predictability based on mobility entropy:Based on the theory of information entropy,the mobile entropy of taxis in Beijing and Portugal Porto is studied.By calculating the three concepts of random entropy,time-uncorrelated entropy and real-entropy,the result gives out the ultimate prediction probability of the two dataset that the predictability of Beijing is between 34%and 50%meanwhile the predictability of Porto is between 35%and 40%.Considering the stability requirement of the mobility prediction for the data,the more concentrated vehicle trajectory of Porto is finally selected for the trajectory prediction algorithm.(3)The third chapter of this thesis focuses on vehicle trajectory prediction.This chapter proposes a new trajectory prediction method based on Diffusion Kernel and gives out simulation result that is better than 1-order Markov method and close to 2-order Markov method meanwhile the time complexity is far lower than 2-order Markov method.According to the simulation results,the accuracy of prediction after adding the one-direction principle has a certain improvement compared to the original method with low accuracy.