Research On Shared Electric Vehicle Scheduling Method Based On Site Demand Prediction

With the development of transportation towards intelligence,sharing,and electrification,shared electric vehicles will become one of the important modes of transportation.Shared electric vehicles are not only comfortable and convenient,but also have the advantages of low-carbon and environmental protection.Sharing can effectively solve the problem of urban traffic congestion.However,due to the tidal and imbalanced nature of user travel,there is an imbalance in vehicle inventory at shared electric vehicle sites.Some sites have insufficient vehicles,resulting in users having no cars available and unable to meet some user needs,reducing the profits of shared electric vehicle operators and hindering the development of the shared electric vehicle industry.In existing research,the scheduling problem of shared electric vehicles is either static scheduling based on historical needs or simplified as a scheduling problem without considering charging,which has limited significance for the development of shared electric vehicles in reality.Therefore,this thesis proposes a scheduling optimization method for shared electric vehicles based on site demand prediction,providing theoretical and practical guidance for the development of the shared electric vehicle industry.Firstly,this thesis preprocessed the original order data of the shared electric vehicle project in Shanghai,and also obtained and processed POI(Point of Interest)data and weather data.Based on the preprocessed travel demand data of shared electric vehicle sites,analyze their spatiotemporal characteristics.The results indicate that the demand for shared electric vehicle travel exhibits strong time dependence and spatial correlation.Compared to weekends,there is more demand for shared electric vehicle travel on weekends,and there is a clear morning and evening peak phenomenon on weekdays.Most users’ car usage time is between30 minutes and 90 minutes.In terms of spatial distribution of travel demand,workdays are mainly distributed in office and residential areas,while weekends are more widely distributed and may also be distributed in leisure and entertainment venues.Then,this thesis studies the demand prediction problem of shared electric vehicles based on sites.Design a new spatio-temporal multi graph convolutional network prediction model based on the analysis of spatio-temporal characteristics.The designed model mainly includes five parts: input layer,LSTM(Long Short-Term Memory)encoding layer,multi-graph convolutional layer,LSTM decoding layer,and output layer.Use LSTM encoder decoder to model temporal dependencies,construct distance maps,interaction maps,and functional similarity maps,and use multi graph convolutional networks to model spatial correlations.By comparing the demand forecasting model proposed in this paper with the HA(Historical Average)model,LSTM model and MGCN(Multi-graph Convolutional Neural Network)model,the model proposed in this thesis has higher accuracy.Finally,this thesis conducts research on shared electric vehicle scheduling based on site demand prediction.This article constructs a mixed integer linear model with the goal of maximizing profits,while satisfying the constraints,and uses an optimization solver to solve it.To verify the effectiveness of the proposed method,this thesis compares the differences between traditional shared vehicle systems,shared electric vehicle systems based on available vehicle scheduling,and shared electric vehicle systems based on site demand prediction.And take the Shanghai shared electric vehicle project as the research object for example analysis.Due to charging scheduling reasons,there are significant differences in scheduling schemes between traditional shared vehicle systems and shared electric vehicle systems that consider charging.Compared to traditional shared vehicle systems and shared electric vehicle systems that schedule based on available vehicles,the shared electric vehicle system proposed in this thesis based on site demand prediction has achieved higher profits.