Research On Travel Demand Characteristics Of Docked Bike Sharing Based On Trip Record Data Mining

Recently,bike sharing services have gradually become an important force on the supply side of urban transportation with the pace of the era of "Internet Plus" and sharing economy.It not only provides a good solution for people’s short-and medium-distance travel,but also can effectively solve many urban diseases caused by urbanization and motorization,which is fully in line with the development concepts of "green travel","carbon neutral" and "peaking carbon dioxide emissions".Generally,bike sharing can be divided into two types in terms of technical characteristics and rental forms:docked bike sharing and dockless bike sharing.Spurred on by the capital,dockless bike sharing development impetus is swift and violent,which influences much on docked bike sharing.The problem of parking chaos caused by the lagging of industrial management has transformed the advantage of parking with docks from an advantage to a liability,causing serious resource misallocation and waste.However,with the stronger stability,controllability,non-profit and other advantages,docked bike sharing is gradually returning to the public view.Meanwhile,with the help of Internet of Things,near-field communication,mobile payment and other information technologies,docked bike sharing is in transition to a smart travel service.The defects in their own design and operation systems are being gradually corrected.Nowadays,"running on docked basis " have become an inevitable direction for the development of the whole bike sharing industry under the requirements of "being ordered".Bike sharing trip record data provides continuous observation ability for large-scale user travel activities,which can effectively reflect the characteristics of user travel demand.However,few domestic and foreign researches have carried out theoretical and empirical research on bike sharing travel demand characteristics based on trip record data mining.Therefore,this study deeply explored the potential value behind the bike sharing trip record data.Then this study combined external environmental data such as built environment and weather information to understand the individual travel demand of bike sharing users and the consequences at the station,network,and system levels.A multi-level research framework was established to systematically analyze the long-term stable characteristics and short-term dynamic change mechanism of bike sharing travel demand,which could help improve the understanding of the internal flow and travel characteristics of urban bike sharing systems.The main research contents of this study are as follows:First of all,the development history of bike sharing and the development history of"running on docked basis" operation of bike sharing under implicit and hardware constraints were summarized.It showed the importance of the different forms of "dock-based" operation patterns in the context of differential fusion management of bike sharing(i.e.docked and dockless).Also,the connotation of this study on the travel demand characteristics of docked bike sharing was defined.Then,the feasibility of urban multi-source traffic big data as a data source for exploring the travel demand of bike sharing was proposed.Second,this study deeply explored the potential value of bike sharing trip record data using data mining and big data analysis.Specifically,this study conducted exploratory spatiotemporal correlation analysis on vehicle numbers,departure and arrival times,and departure and arrival stations between trip records.After establishing association rules based on spatiotemporal matching and reconstructing the full life cycle activity chain of each shared bicycle,this study could extract the activity status of shared bicycles such as being used and scheduled;Then,given that the movement of shared bicycles was conducted based on the stations,the operational characteristics of the site level are also closely related to the travel needs of users.After connecting the activity status of each shared bicycle to the corresponding bike sharing station,this study defined two evaluation indexes:station vitality(SV)and station pattern(SP),which can basically summarize the operating characteristics of stations.Accordingly,there were seveal bike sharing stations’ operating performance could be identified based on K-means++method;At the network level,using the data of shared bicycle being used and scheduled,this study constructed complex bike sharing users-riding complex network and system-scheduling complex network based on graph theory and adjacency matrix.Then,this study evaluated the operation status and user demand characteristics of the bike sharing system through the overall topology analysis.Third,the interaction relationship and influence mechanism between the built environment and the bike sharing travel demand were systematically discussed.Firstly,bike sharing stations were classified based on the built environment in the station coverage area to reveal the possible similar urban spatial layout and spatial functional structure.On this basis,the dynamic topic model(DTM)and gradient regression decision tree model(GBDT)and multi-scale geographic time-weighted regression model(MGTWR)were constructed from the disaggregated and aggregated perspectives respectively.Specifically,as for DTM,based on the large-scale bike sharing trip record data and POI data,the riding records were linked with built environment characteristics of the corresponding stations.Then,each bike sharing riding record could be converted into a text and would be an input to the DTM.There were several key bike sharing trip patterns and trending could be identified.Notably,relative to the average effect mapped by the fixed coefficients of the linear regression model,GBDT could be used to test the complex nonlinear effects and possible threshold effects between the built environment factors and bike sharing stating returns.Notably,MGTWR solved the problem of temporal and spatial heterogeneity,and could also address the different temporal and spatial heterogeneity of various built environment factors.Fourth,this study explored the short-term travel demand forecast of urban bike sharing based on the deep spatiotemporal learning method.Based on the multi-source datasets including bike sharing trip record data,built environment data and weather information data,a GraphSAGE-LSTM network that suit the short-term prediction of the borrowing and returning demand of bike sharing stations was constructed.Combined with the respective advantages of GraphSAGE and LSTM algorithms,GraphSAGE-LSTM model could simultaneously extract the temporal dependence and spatial correlation of ridership changes,which greatly improved the accuracy of short-term travel demand prediction.Besides,this study compared the forecasting performance of the GraphSAGE-LSTM model and other baseline models under different forecasting time intervals to verify the excellent performance of the proposed model.Finally,the study further applied the multi-level analysis framework of the travel demand characteristics of bike sharing on domestic data,and selected Wenzhou as the case city.Firstly,the riding records and scheduling records of bike sharing were extracted by using the bike sharing trip record data,built environment data and weather data,and then the relationship characteristics and interaction mechanism between the bike sharing travel demand and system operation were summarized in terms of station operation and network structure.Then,the influence mechanism of the built environment behind the different spatiotemporal travel patterns of the individual travel behavior and station-level aggregated ridership was quantitatively analyzed.It helps us to raise awareness about the spatiotemporal variation law of macro and micro bike sharing travel demand from the perspectives of non-aggregation and aggregation.Finally,the trained deep spatiotemporal learning model was used to predict the short-term travel demand of bike sharing,and its spatiotemporal evolution law was discussed.