Research On Methods Of Urban Public Bicycle Sites Demand Prediction And Scheduling Optimization

Urban public bicycle,which is the product of culture-enriched,technology-empowered and environment-friendly transport in new age,has three main functions:Last mile of mass transit,serving for short-distance travel and serving for scenic spots and campus.In face of the common phenomenon that no bicycle can be borrowed or no position is vacant for returning a bicycle at sites in many cities,it is found that inefficient schedule is the main reason.So,it is of great significance that how to predict precisely the travel demand at each site in advance and arrange reasonable scheduling path to solve the problem.As the PBS(public bicycle system)operation data accessible of domestic cities is less,this article does research on PBS travel demand and scheduling optimization based on bicycle travel data of New York City.Through the analysis of data in New York City,we find both in China and New York City,public bicycle demand has morning and evening peak on weekday.Furthermore,only in June 2016,there were 1.46 million reliable data of public bicycle travel which contributes a lot to the analysis of the public bicycle operation.So the predicting method based on travel data in New York City is also applicable to domestic cities.The specific research work in this paper can be divided into 4 parts as follows:Firstly,this paper describes the position of PBS in public travel system and analyzes the subjective and objective factors affecting travel demand based on 1.46 million public bicycle trips in New York in June,2016,finding that travel rules on the weekend are obviously different with that on weekday.Secondly,this paper introduces the theory of BP neural network,and proposes a new method to predict the demand of bicycle rental considering the date attribute.When predicting the travel demand at each site by using of BP neural network,the date attribute of the input variables should be consistent with the output variables,which means,they are all weekdays or all weekends.This paper makes a prediction based on the data by using SQL Server Database and Matlab which results that the relative errors are low and the minimum relative error is 1%.Thirdly,this paper summarizes the definition,classification and optimization algorithm of VRP(vehicle scheduling problem)and give a two-step processing of solving VRP called scheduling area division and scheduling in each area;Firstly,this paper put forward a method to divide all sites into multiple zones by clustering three times based on combining association rules with cluster analysis.Secondly,a scheduling model is set up for the sites which have scheduling requirements in each area,and the goal is to minimize the number of vehicles and the total scheduling time.The ant colony algorithm with 2-opt local search is used to find the optimal solution.Finally,this paper performs a simulation of scheduling path optimization based on data of 21 public bicycle station by MapGIS Tool for latitude and longitude conversion and Matlab for main process.By simulation,we find that the ant colony algorithm with 2-opt local search performs better and makes scheduling time more shorter than the conventional one.