Research On Demand Prediction And Scheduling Optimization Of Shared Bikes

Bike sharing is an innovation of "Internet +",which solves the problem of "last 1kilometer",facilitates our travel,and effectively alleviate traffic congestion and air pollution.However,the demand for shared bicycles will continue to change with time and location,so,the phenomenons of not being able to find shared bicycles or parking a large number of shared bicycles in a certain place often occur.In order to solve the problem of mismatch between the demand and supply of shared bicycles in time and space,based on the order data of Beijing Mobike in May 10-16 2017,this paper buildis a shared bicycles demand forecasting model and a scheduling model.The demand forecasting model is oriented by location,time and weather.The scheduling model is oriented by the difference between the demand and supply and the driving distance of dispatched vehicles.First of all,this paper determines the factors which affect the demand of shared bicycles through the travel feature analysis of users.Feature analysis includes time and space.Time characteristics is analyzed from two perspectives which are hour and whether it is weekend.Space characteristics include the riding distance,the start and end distribution area of the shared bicycles in the morning and evening peaks.The results found that hour,whether it is weekend and location have a greater impact on the demand of shared bicycles.Secondly,based on the result of feature analysis,POI analysis and literature summary,the independent variables for the prediction of the amount of borrowed and returned shared bicycles are determined,and integration algorithms are used to predict,so as to determine the scheduling demand.This paper mainly uses random forest in bagging ensemble algorithm,XGBOOST and LightGBM in boosting ensemble algorithm,and stacking ensemble algorithm which is built by BP neural network,decision tree,multiple linear regression and stacking ensemble principles.The results found that among the three categories of integration algorithms,the best prediction performance is the boosting integrated algorithm.Among them,RMSE is the minimum when LightGBM predicts the number of borrowed bikes and XGBOOST predicts the number of returned bikes,which are 4.2840 and 5.4312,respectively.Finally,based on the predicted number of borrowed and returned shared bicycles,the number of shared bicycles that need to be transferred in or can be transferred out of each station can be determined,and on this basis,the objective value can be calculated.And it can determin the order in which the stations are serviced.Based on the location of stations and the cumulative number of shared bicycles that need to be transferred in,the dispatching center is determined,so that the starting and ending points of each dispatched vehicle are all the dispatching center.Aiming at the minimum of the difference between the demand and supply and the driving distance of dispatched vehicles,scheduling model of shared bicycles which integrates pickup and delivery is established.The mutation and crossover operations in the genetic algorithm are added to the particle swarm algorithm to form a hybrid particle swarm algorithm.And it is used to solve the scheduling model and determin the order in which the stations are serviced.The results show that in the process of solving the scheduling model,the value of the objective function decreases rapidly,therefore,this algorithm can quickly solve the model.