Research On Multi-factor Forecasting Model And Scheduling Algorithm Of Public Bicycle Demand

With the rapid development of society,people are now faced with ever-accelerating pace of rhythm of life and work.The number of motor vehicles is constantly rising,causing the traffic congestion,energy shortage.As an innovation of urban public transport,public bikes effectively meet the need of people for green travel.However,the unbalanced distribution of public bikes is one of the most difficult problems,which greatly affects the user experience.Therefore,in this paper we analyze the relevant factors which affect public bike usage,forecast the usage pattern and optimize public bike dispatching,which provides an important theoretical basis for the optimization of public bicycle system.Firstly,we introduce the research status of public bike station planning,demand forecasting and bike dispatching,then analyze the existing problems of imbalanced demand and large dispatching costs in public bike-sharing system.Besides,we describe the research methods and related theories,including historical data mean model(HM),Auto-Regressive and Moving Average Model(ARMA),as well as some heuristic algorithms used to solve the vehicle routing problem.Secondly,according to the travel data of New York public bike-sharing system named Citi-Bike,and the historical weather data,we propose some relevant factors that affect the demand of public bike,and analyze the time factor,location factor,weather factor and temperature factor respectively.Then,the weighted K-means is used to cluster the stations,and the lagged variable model is proposed to simulate the impact of the weather on the demand.We propose multiple factor regression model with ARMA error(MFR-ARMA)to predict the demand of each cluster.The experiment results show that the average error rate of MFR-ARMA is 28%,which is about 8%better than the prediction model of historical mean and ARMA.Finally,we build a constrained public bike dispatching optimization model and analyze the difference between the problem based on actual distance and based on Euclidean distance.In order to minimize the total actual distance of the dispatching path,we propose the genetic algorithm based on 2-opt optimization,and analyze the feasibility of the algorithm to solve the actual distance dispatching optimization problem.The experiment results show that the average distance of the optimal solution for the algorithm is 38.02km,which is 3.3%better than that of the traditional genetic algorithm.At the same time,its convergence speed is 61.1%higher than that of the traditional genetic algorithm.