Traffic Flow Prediction And Rental Location Of Bike-sharing System

Internet rental bicycle(hereinafter referred to as bike-sharing)came into being in 2014,which has achieved rapid development in a short time.Bike-sharing not only promotes the prosperity and development of the sharing economy,but also become one of the indispensable green travel modes in city.As one of the best ways to connect short distance travel and public transport,bike-sharing greatly facilitates people's travel.In this paper,the short-term traffic volume prediction of bike-sharing and the location of recommended parking spots are two research directions.The short-term traffic volume prediction of shared bicycles in the region can not only grasp the travel demand of bike-sharing in the region in time,but also predict the peak travel time and hot areas,so as to facilitate the subsequent scheduling of bike-sharing.The research on the location of bike-sharing parking spots is helpful for enterprises to reasonably plan and build electronic fences,guide and standardize users' parking behaviors,and avoid the phenomenon of affecting the city appearance and occupying other road resources.The research of these two directions is very important theoretical and practical significance.Firstly,this paper analyzes the travel characteristics of bike-sharing from four aspects: time characteristics,distance characteristics,spatial characteristics,and riding section preference.The results show that there are two peaks in the travel time distribution of bike-sharing.The activity of bike-sharing in Beijing urban area is higher than that in suburban area,and the cycling frequency is higher in residential area,work area and the surrounding area of subway.Bike-sharing is mainly for short-distance travel,and the cycling section is mainly concentrated in the branch roads and alleys outside the main roads.Secondly,a GCN-LSTM short-term traffic prediction model which based on deep learning is proposed.The map of Beijing is divided into rectangular grids,and the grid area is regarded as the bike-sharing virtual station.The spatial relationship between the stations is extracted by using the convolution neural network,and the travel time characteristics of the bike-sharing are extracted by using the LSTM model.In the case,Geohash coding method is used to divide Beijing into grid like areas.The travel volume of the station and the travel records across the grid area are counted in 30 minutes.Distance map and interaction map are established to extract the spatial information between adjacent blocks,and then they are input into the LSTM model to extract the information of time series.The short-term traffic of bike-sharing in the designated area is analyzed Volume to forecast.The experimental results show that the GCN-LSTM model proposed in this paper can effectively capture the spatial correlation and time dynamics of bikesharing,which also can reduce the prediction error.Finally,from the perspective of the interests of bike-sharing enterprises and users,a bilevel programming model for the location of recommended parking spots of bike-sharing is established at the street level of the community.The upper level organization establishes the total cost model of bike sharing enterprises from the perspective of bike-sharing enterprises.The lower level organization establishes the time satisfaction model and the availability model of bike sharing from the perspective of bike-sharing users,The NSGA-2 algorithm is designed to solve the problem and get the optimal location scheme and the number of parking spots.In this case,the Jiugong area of Beijing is selected,and the optimal location and vehicle size allocation are obtained by using this model,which verifies the feasibility of the model proposed in this paper.