Taxi Pick-up Stations Recommendation Based On Grid

At present,many cities in our country are faced with the problem of traffic congestion,urban residents travel becomes more and more difficult,How to improve the relationship between taxi and passenger supply and demand is an urgent problem to be solved.Taxi trajectory data contain a wealth of information such as temporal and spatial attributes.By mining the knowledge hidden in these data,you can provide location-based services to the taxi,make people commuter more convenient,so that the city’s operation will be more reasonable.This paper makes use of 4000 taxitrajectory data in a month in Shanghai,mainly completed the following contents: 1)data processing,through data deduplication,data cleaning,data reordered and other pre-processing operation,obtain all the pick-up location points data within the selected region,and the map is divided into 200m×200m grid.Then the pick-up location points data are mapped to the corresponding grid,the grid based pick-up points is the basis of this paper;2)distribution of pick-up points model,analysis of residents commuter.we proposed a Grid based Gaussian Mixture Model(GGMM)with temporal and spatial dimentions unified that groups the data into a number of temporal-spatial clusters by observing the data at different time of the day in each grid cell,through comparing with GMM and PMM algorithms,the result show that the model can effectively describe the pick-up point distribution,for different Gaussian number(5-30),in the Perplexty index,GGMM reduce 6.5%-27.1% compared with GMM;3)prediction of the number of the pick-up points based on grid,an improved ARIMA model is proposed to set up time series.The experimental results show that the model is accurate and has higher short-term and long-term prediction ability;4)recmmender of taxi pick-up locations,for the recommendation system often only consider the capacity of each location,but did not consider the driver’s personal preference,respectively analyzed collaborative filtering based recommendation algorithm and latent factor model,at the same time,added the time factor into CF model,and that makes a higher accuracy rate.When the recommended number is 2,the accuracy is increased by nearly 10%.and then used LFM,the accuracy of this method is proved to be the highest,and the accuracy of the method is about 4% higher than the CF method.at last,proposed a mixed model that combined with the predicted pick-up number and the driver’s preference obtained from the recommendation algorithm,then the obtained pick-up locations from this model is recommended to the driver.