Research And Application Of Traffic Prediction And Rebalancing Method For Shared Bikes

Shared bikes have been widely used in many large cities,providing users with a convenient choice for the "last mile".However,bike usage in a city is often uneven and "tide phenomenon" occurs frequently.There are some bikes piled up in some stations,and there are no bikes available in some stations.How to effectively predict the bike traffic at each station and rebalance the bikes at each station to reduce user churn is a problem that needs to be resolved.Massive user cycling data provides an opportunity for traffic prediction and rebalancing management of shared bikes.It enables bike-sharing companies to save resources on the premise of improving operating efficiency,thus facilitating the construction of "smart city".Therefore,traffic prediction and rebalancing for shared bikes has become a hot topic in academia and industry.Part of the research regards the effect of the user’s historical behavior on the future as consistent one and ignores the different effects of the user’s historical behavior on the future behavior in the process of changing with time.In addition,some studies have been conducted to incentivize users or set rules for bike rebalancing,which is not efficient.In order to solve the above problems,this thesis proposes a traffic prediction model of shared bikes,which is based on long-short term memory network and attention mechanism.We study the rebalancing problem of shared bikes based on q-learning algorithm in reinforcement learning.The main work of this thesis is as follows:(1)This thesis analyzes the research background,research status,relevant theories and shortcomings of the current prediction and rebalancing algorithms for shared bikes.(2)This thesis presents a shared bike traffic prediction algorithm based on LSTM network and attention mechanism,which combines cycling context information.The data set of shared bikes is analyzed in detail,and in view of the problem that most of the current traffic prediction focuses on the characteristics of time dimension and ignores other influencing factors,the cycling context information is integrated.At the same time,a method of traffic prediction based on long-short term memory network and attention mechanism is proposed.The model effect is verified on real data set.(3)This thesis proposes a rebalancing algorithm based on reinforcement learning.The clustering algorithm is used to divide the urban bike-sharing stations into regions.On the premise of meeting the regional rebalancing needs,a bike-sharing rebalancing algorithm based on reinforcement learning is proposed to assist the bike operation manager to make scientific rebalancing decisions.(4)This thesis designs and implements the prototype system of traffic prediction and rebalancing management of shared bikes.Based on the algorithm proposed in the thesis,the traffic prediction and rebalancing management system of shared bikes is designed.The system can provide detailed overview of users’ historical cycling data for bike-sharing operation managers,who can query the traffic prediction results of any station and dispatch bikes according to the given suggestions.