Snatiotemporal Data Mining And Modelinig Of The Service Supply,Flow Network And Travel Demand Of The Public Bicvcle Svstem From The Perspective Of Urban Computing

As an important component of urban non-motorized traffic,public bicycle system plays an important role in alleviating traffic congestion and improving commuting efficiency of urban residents.The concept of green-travel from public bicycle system can also help reduce environmental problems such as greenhouse gas emissions,noise pollution and fossil energy consumption.In the past decade,Chinese cities following Hangzhou have speeded up the construction of the public bicycle system.However,due to the lack of theoretical research on the demand modeling and forecast of residents' short distance travel,the imbalance between the supply and the demand as well as the idle problem of the public bicycle system have weakened the utilization rate and raised the unsatisfaction of citizens.At the same time,the development of social sensing and urban computing technology in the era of big data makes it possible to grasp the pulse of urban development and makes it a new means of modern urban planning,management and solving urban problems.Therefore,in view of the lack of studies on travel demand modeling and understanding of the dynamic travel pattern of public bicycle system,this paper uses the geographic analysis,network analysis and machine learning as research methods,the Hangzhou Public Bicycle System as the research object,city open data as data sources,to complete three tasks:uncovering the existing travel demand structural and identifying the demand conflict area of interesting;predicting the interaction flow from station-based data;and building the demand forecasting model based on urban multi-functions.The paper contributes to urban short-distance travel theories from service supply-network flow-traffic demand analysis aspect and provides theory and method support of the public bicycle systems demand modeling.The main research contents and conclusions of this paper are summarized as follows:(1)Research based on the theory of travel demand and the methodology of spatiotemporal analysis and modeling,compared the differences and similarities between Chinese and western countries public bike-sharing systems with respect to the traffic facilities supply base,traffic role in the peak time,residents' travel preference and so on,and set up a data mining modeling framework on the of service supply,flow network and travel demand spatio-temporal,which added new findings to the existing urban public bicycle travel theory.(2)Considering the two-side opposite or staggered layout of Chinese public bike-sharing systems,a sub-nearest neighbor strategy is designed to calculate the station spacing and can more accurately express the spatial distribution characteristics of the system than the nearest neighbor analysis.What's more,by introducing the concept of spatial proximity into the definition of demand conflict,about 90%of pseudo-demand conflict stations can be filtered out,so as to more accurately identify the demand conflict area of interesting(AOI)and guide the operation in the bicycle balancing.(3)Based on the sparsity of strong interactive pairs between public bicycle stations,the paper designed a sparse representation of the transformation model between the station-based data and the trip data,extracted the distance preference and the direction preference of short distance travel,and realized an inferential model for the dominant flow pairs.In the experiment of Binjiang district,the model accurately reveals the strong interactive flows in morning and evening rush hours,which shows that the public bicycle stystems can extend inter-city commuting by extending the accessibility of the existing urban transport systems(subway and bus)and convert walking to complete the intra-city commuting.(4)The paper comes up with a two-stage forecasting model for the travel meand of public bike-sharing systems.Combining multi-source city open data and using the topic modeling technique in text mining,functional zones and compound features are extracted and identified from multi-source and unstructured mobility data.On this basis,together with features extracted from the urban structure and short-distance mobility,the demand model is constructed by using supervised learning technology to fully explore the correlation between the demand model and urban functions.The experimental results in the downtown area show the priority of the model over models using land use types or the land use diversity index,and its strong expansibility.This study provides a large number of empirical research conclusions on the spatial and temporal travel pattern of public bicycle systems in Chinese cities,which can provide references for other relevant studies.Besides,the study further confirmed the close ties between short-distance travel demand with the urban functions and the urban layout,and provides a feasible prediction model,which improves the modeling accuracy,enrichs and promotes the application of intelligent traffic and urban computing technique in smart cities.