Research On Key Technologies Of Scheduling Demand Prediction And User Diversion Of Bike-sharing

With the development of the public bikes provided by government and the stationless shared bikes provided by companies,they connect the "last mile" between the riding origin and the destination position of users.They are both green and environmentally friendly way of traveling.But in recent years,there are many problems appeared in the operation processes.Due to the existence of tides,public bikes are difficult to rent or return and the station-less shared bikes are always no bikes that can be borrowed(or only bad bikes)when users are renting.The chaos parking is also a problem.The unbalance of traffic flow not only causes the decline of bicycle utilization,but also causes traffic congestion and urban management confusion.In order to solve these problems of sharing bicycles effectively,this paper considers the combination of two modes of bike-sharing: public bikes and shared bikes and proposes new solutions from two aspects: dispatching demand prediction and user diversion.And the machine learning technologies are used to predict more intelligently.Then the paper firstly introduces the similarities and differences between public bikes and shared bikes,the development status at home and abroad and then introduced the related research at home and abroad,proposes two kinds of operation problems of lack of sharing and research of bikes.Then this paper uses the public bike data of Hangzhou city as an example to analysis the influence factors of bicycle traffic sharing,then defined the variation values of traffic flow and the method virtual site of sharing bikes based on Geo Hash is also introduced.Then a new machine learning model called SMVP based on the stacking ensemble XGBoost to predict the values is proposed.The experiments are done based on Hangzhou public bicycle and New York City public bicycle real-world datasets.This proved that the accuracy and reliability of the prediction model.In addition,this paper put forward two methods of regional dispatching demand prediction,and then made some analysis of regional demand balance.Combined with the parameters of regional traffic volume,bike volume and threshold,a new calculation method is proposed.Then,based on real-world datasets of Mobike shared bikes users,the cycling travel regularities are analyzed.The user candidate prediction model CGM based on FP-Tree is proposed.After predicting the candidate in the future,a deep learning model called DPNNst based on a combination of Long and Short-Term Memory network(LSTM),Convolutional Neural Network(CNN),Fully Connected Neural Network(FC)and other neural networks to predict travel destination of the user is proposed.Then the experiments on real-world datasets of Mobike Beijing to validate the model effect are well done.After that,the user diversion strategy of shared bicycles is studied,and the guiding way of sharing bicycle users is put forward,such as,recommending return bikes points,guiding users to assist diversion,and guiding traffic pressure mitigation.Finally,this work summarizes the methods of bicycle sharing traffic prediction,demand prediction,travel destination prediction,user diversion and so on,and looks forward to the future development and research trend of bicycle sharing.