Demand Hotspot Extraction,Forecasting And Rebalancing Of Bike Sharing Based On Spatial Crowdsensing Big Data

Due to the uneven spatio-temporal distribution of user’s travel demand and the limited capacity of bike sharing parking area,the contradiction between supply and demand of bike sharing system has always been the key factor hindering the development of the system,whether it is the traditional bike sharing with the dock or new bike sharing without the dock.To solve the contradiction between the supply and demand of bike sharing,an effective method to rebalance bikes in the bike sharing system is in great need.However,the existing bike rebalancing methods have problems such as incomplete demand hotspot modeling,inaccurate demand forecasting,and unreasonable rebalancing strategies design.To this end,this thesis studies the bike demand hotspot extraction,forecasting,and rebalancing based on bike sharing crowdsensing big data.The contributions of this thesis are as follows:(1)We propose a spatio-temporal semantic extraction method of bike sharing demand hotspots based on clustering and peak detection algorithms.This method first extracts hotspot spatial regions based on a distance-constrained clustering algorithm and then extracts hotspot temporal trends based on a peak detection algorithm.This method solves the problem of existing methods in extracting demand hotspots and is more helpful to the demand forecasting model design.(2)By jointly modeling the spatial,temporal,and contextual features,we propose a hotspot demand trend forecasting model.Firstly,the model constructs spatial neighborhood graph,spatial pattern graph,and spatial function graph to model multiple spatial relationships,and fuses multiple spatial features based on multi-graph convolution network.Secondly,the model is based on long short-term memory and autoencoder to model temporal and contextual features.Finally,by fusing the spatio-temporal contextual features,the demand hotspot trend forecasting model is proposed.This model can accurately forecast the trend of demand in the future.(3)By modeling the bike rebalancing task as a spatial crowdsensing task,we propose a bike sharing demand hotspot rebalancing method based on integer linear programming and temporal difference algorithm,which achieves effective bike rebalancing and longterm incentive for users.Firstly,the rebalancing tasks are generated based on the integer linear programming,and the incentive mechanism of rebalancing participants is designed.Secondly,the rebalancing tasks are allocated based on the temporal difference algorithm to maximize the rewards of users participating in the bike rebalancing.This method can optimize the effectiveness of user-based bike rebalancing.The experimental results on two bike sharing datasets show that our proposed method extracts,forecasts,and rebalances bike demand hotspots comprehensively and effectively.This method is helpful to solve the contradiction between the supply and demand and improve the service quality of bike sharing system.