Research On Repositioning Demands Of Free Floating Bike Sharing Based On Travel Data Mining

With the diversified travel demand of residents and the deepening of the concept of “Internet plus”,sharing transportation represented by bike sharing has become a key direction for urban transportation development.Free floating bike sharing has risen rapidly because of the convenience and fashion,but its barbaric expansion has also caused widespread social worries,especially the chaotic having unrestrained impact on the traffic order.Based on the Nanjing travel order data provided by Mobike,this study deeply explores the characteristics of bike sharing travel,analyzes the travel demand through various clustering methods to determine the virtual stations,uses a variety of models to predict the short-term passenger flow,and finally realizes the accurate identification of bike repositioning demands for the virtual station service area.This paper firstly analyzes the basic situation of bike sharing and lays a foundation for subsequent research.From the aspects of technical characteristics,current problems,local management policies,etc.,we briefly introduced the overview of bike sharing and made recommendations for the future development direction of the industry.A preliminary analysis of the Mobike order data in Nanjing was conducted,assessing business operations with the scope of service,fleet size,number of users,number of travel orders,bike turnover rate and other indicators to,exploring user demand with the frequency of use,travel time,travel distance,etc.Effects of weather and metro travel demands are considered.Then,based on the clustering of travel demand,a virtual station identification method is proposed.Cluster analysis is carried out on the spatio-temporal characteristics of bike sharing travel demand,and the cluster center is determined as a virtual station,and the cluster boundary is used as a virtual station service area.The K-means clustering,density-based clustering and spatio-temporal clustering methods are analyzed,and the contour coefficient,CH index and station radius are use to evaluate the results,so the reasonable and reliable virtual station recognition results are obtained.The analysis shows that the K-means clustering results are the best,and there are about 4,000 bike sharing virtual stations in the Nanjing built-up area.Then,using a variety of models for short-term forecasting of passenger flow,it provides the possibility for real-time dynamic identification of repositioning demands.Linear methods are historical average and time series,nonlinear methods are neural network use and random forests.The mean absolute error,root mean square error,and R-square are used to evaluate the pros and cons of various prediction methods.It is found that the random forest prediction results are the best,but the other three methods can also achieve accurate predictions.Finally,the virtual station repositioning demand identification method is established with reference to the actual operation of the enterprise.The analysis of operational data shows that the repositioning capacity of Nanjing Mobike is about 20-30 thousand bikes per day.The number of bikes in the area needs to be kept within a reasonable threshold.The upper limit is to ensure that the bike density can not be too large to avoid nowhere for parking.The lower threshold is to ensure that the number of bikes can not be too small to avoid no bike for using.The repositioning demands are effectively identified based on bike threshold and passenger flow prediction,and a practical integrated repositioning method is proposed.