Research On Taxi Recommendation System Based On Gps Trajectory Data

In recent years,with the continuous improvement of material life level,people’s traveling demand is rising sharply,which results in the problem that it is very difficult to hail a taxi in the city.Although the number of urban taxies is increasing constantly,the empty loading rate of many taxies still remains high,so how to effectively improve the supply-demand relationship between passengers and taxies has become a serious problem.On the other hand,most taxies in the city have equipped with a GPS device currently,so massive GPS trajectory data is sent to the dispatch center every day,which contains passengers’ mobility and taxies’ pick-up behavior information and provides a great advantage for the taxi recommendation.Taking this as the starting point,this thesis first gets all the pick-up points from the mining results of a large number of GPS trajectory data,and then establishes a recommendation forecasting model,which can be used to complete the recommendation for taxi drivers and passengers by some certain policies.This thesis focuses on the solving of the supply-demand imbalance problem between taxies and passengers,which has both theoretical and practical values.The main works in the thesis includes:(1)Preprocessing of taxi trajectory data.Firstly,we introduced the original data set used in this thesis and formulated some special rules for the data set according to our need,and at the same time,we proposed a specific method to obtain the pick-up points of taxies.Then map matching algorithms,which were used for the preprocessing of data set,and specifically a map matching algorithm(IVMM)used for GPS trajectory with low sampling rate were studied and analyzed.Finally,in order to make the preprocessing more efficient,a distributed data preprocessing scheme was given on the Hadoop platform.(2)Cluster Analysis of taxi pick-up points candidate set.Firstly,we made a research about the existing clustering algorithm,especially the K-Means and DBSCAN algorithm.Then we analyzed the advantages and disadvantages of these two algorithms in dealing with the GPS trajectory data set.Combining the advantages of these two algorithms,we proposed a new algorithm based on them.Through experiments and analysis of the results,the effectiveness of the algorithm was validated.(3)Research on taxi recommendation method and implementation of the prototype system.Through further statistical study on historical trajectory data,this thesis established a recommendation and forecasting model and proposed a method used for the calculating of passenger-carry probability,income estimation and waiting time.On this basis,combined with the "cost-risk-benefit" analysis method,we designed a set of complete recommendation strategies and illustrated all the recommended process steps.Finally,we developed a prototype system of taxi recommendation based on all the previous works.