Research On Rebalancing Of Public Bicycle Based On Neural Network

As road traffic conflicts worsen,countries around the world encourage people to use green and low-carbon modes of travel.This paper combs the relevant literatures of public bicycle systems at home and abroad,expounds the background of the public bicycle system and analyzes user characteristics.According to the total travel data of urban residents in New York,USA,this paper analyzes the subjective and objective factors that influence the travel demand.Aiming at the phenomenon of "rental and returning the bike" in the bicycle sharing system,this paper researches the main problem in the operation process that are unreasonable scheduling,not timely,etc.According to the urgency of scheduling demand of each station,use the scheduling vehicle to schedule the bicycle from the surplus rental point to a station with insufficient stock according to certain route requirements.The bicycle storage capacity of each station is within a reasonable range,satisfying the demand of "having a bike to borrow,and having a place to return",making the rental system healthy and stable development.Firstly,this paper forecasts the rent and return of each rental point.BP neural network is used to fully study the historical data of renting and returning bikes in New York in the United States in 2018,training the forecasting network,and comparing the forecasted data with the test set data to test and verify the accuracy of the training network,so as to predict the amount of rented bikes in the future.Then build the demand forecasting model,comprehensively consider the bicycle inventory of the site and the amount of bicycle borrowing and returning bikes in the future,and determine the final total scheduling demand.Secondly,the scheduling balance is performed for each rental point.It includes two aspects:partitioning of scheduling area and scheduling routing optimization.Firstly,the K-means algorithm is used to divide the sites according to the distance.Then,by mining the historical group data between the stations,the mutual connection between the stations is found,according to the complementary relationship of the bikes between the scheduling area.The concept of mutual balance is proposed,and the sites is divided into two by the mutual balance degree.The division of the scheduling area considers both the distance and the scheduling quantity,which makes the division of the scheduling area more reasonable.In each scheduling zone,regardless of the change of the site inventory with time,the static scheduling model is established to optimize the driving route of the scheduling vehicle.Under the requirement of the vehicle capacity limitation,the total route length of the scheduling vehicle is the shortest.This chapter adopts SOM neural network method constructs an initial petal of K nodes of K rings,and randomizes the scheduling order of each station.At each station,surrounding nodes generate a winning node through competition,and determine the position of the ring where the winning node is located.The location of the station on the scheduling route,and continuously update the competing node and its neighbor nodes,and continuously compete until the distance between the node and the station is less than the limit value or the maximum number of iterations is reached,the algorithm stops,and finally the scheduling route is determined.Finally,based on the actual operational data,the optimization and simulation of scheduling rebalancing for 23 bicycle stations in New York,USA.Comparing the results of the modified SOM algorithm with the results of the simulated annealing algorithm,it is proved that the total path distance obtained by the modified SOM algorithm is shorter and the algorithm is more efficient.