Short-term Demand Forecasting Of Network Car Based On Spatio-temporal Feature Analysis

With the development of network technology and the popularization of intelligent terminals,online car-hailing has gradually become the preferred way for people to travel by characteristics of speediness,convenience and high-quality services.However,the phenomenon of no-load or shortage of supply is still very serious in real life.With the increasing popularity of online car-hailing service such as Uber and Didi,we can continuously collect large-scale demand data for online car-hailing.How to use these big data to change demand forecasting is a key practical problem.Therefore,this paper uses the internal data and external data collected on the network to analyze the spatial and temporal characteristics of the demand and to forecast the short-term demand for the online car-hailing.Firstly,this paper studies the order data and GPS trajectory data,analyzes their characteristics,and complete data preprocessing.Then,based on the data of the demand of the online car-hailing,analyze the time characteristics of the travel demand,divide the different time period types,and obtain difference between the demand of the working day and non-working day.On this basis,analyze the spatial characteristics of travel demand of online car-hailing in different periods,which provides basis for short-term demand forecasting.In order to accurately capture the temporal and spatial characteristics of the short-term demand of online car-hailing,a CNN_LSTM_Kalman combined forecasting model is constructed based on considering various factors.The model has three views(the correlation between future demand value and near-time point by is modeled by LSTM),the spatial view(local spatiality is modeled by CNN)and revised view(reduce the errors between forecast and reality).Through verification,the prediction accuracy of combined forecasting model is higher than that of single model(LSTM forecasting model,Kalman forecasting model).Finally,taking the second ring road area of Chengdu as an example,this paper applies the validated combined forecasting model to the actual forecasting,and puts forward relevant suggestions for the online car-hailing reservation platform,traffic operation management,driver and passenger respectively.The results of this paper can provide reasonable vehicle layout basis for the online car-hailing drivers or related companies.From the perspective of the online car-hailing drivers,the forecast of vehicle demand can make them know the passengers demand in different areas beforehand,and then they can go to the areas with more demands,thus avoiding the blindness of searching for passengers.From the perspective of the operation and management of the online car-hailing of the city,the forecast of the online car-hailing demand can get the passengers demands in different areas in short time,thus triggering forward-looking matching and scheduling behavior.