Application And Research Of Traffic Congestion Diffusion In Metro-city Scales Based On Multi Source Data Fusion

With the development of urbanization,daily travel requirements increase year by year.However,the lumbering road resource growth has become increasingly unable to meet the rapid growth of travel request.Consequently,the problem of traffic congestion diffusion has bacome one of the major issues for cities.However,at the same time,people are bothered by the complex traffic road and serious traffic jam,also,great economic loss is due to the inefficient traffic.Hence,it is significant to develop a traffic congestion diffusion model under complex urban traffic environmentHowever,the urban tranport system i5 eomplex and diverse and urban traffie con-dition is influenced by many factors,which make it essential to have a comprehensive understanding of the whole city from multiple perspectives.And fortunately,multi-source data make it possible to study traffic congestion diffusion in a metro-city scales Because urban traffic data hsa the problem of the data spasity,data loss,data bias and data heterogeneity,the esisting data fusion methods applied in ubran computing are often used in the data preprocessing,correlation analysis or direct fusion stage.They ignore the fact that all influence factors on traffic condition are changing both in spa-tial domain and temporal domain.This dissertation utilizes a multi-source data fusion method to study the traffic congestion diffusion phenomenon under metro-city scales The novel contributions are summarized as follows1.Propose a data fusion method for traffic condition predieiton by using homogeneous data based on weight assignment method:To address the problem of the data spasity,data loss and data bias for GPS data,this dissertation contracts a data fusion method for traffic condition prediciton by using homogeneous data based on weight assignment method.A dynamic weight assignment mechanism is designed according to the quantity and quality of each data type to fulfill the missing data2.Propose a data fusion framework for homogeneous data based on branching-transform-fusion strategy:To address the data heterogeneity problem,this disserta-tion puts forward a data fusion framework for homogeneous data based on branching-transform-fusion strategy.With different data type inputing different neural network,the spatiotemporal representation for each data is extracted by its neural network followd by a fusion operation in feature level.The metods of embedding and parameter sharing are widely used in this model,which makes our model scalable and easy to train3.Prosed a traffic congestion diffusion model by using multi-source data fusion:To define traffic congestion under complex urban transport system with hetero-geneous roads,this dissertation propose a traffic congestion diffusion model by using multi-source data fusion.In the model,the definition of traffic congestion is derived from large urban data.As far as we know,we are the first to study from the perspec-tive of traffic flow influence in metro-city scalse,.A traffic congestion state evolving process coupled with crowdedness updating is designed to characterize the process of congestion formation,diffusion and dissipation4.Propose a traffic bottlenecks identification model from influece maximizaiton perspective:To find the key point of urban traffic dynamiclly and globally,this dis-sertation proposes a traffic bottlenecks identification model from influece maximizaiton perspective.We extend the scope of traffic bottlenecks definition based on traffic con-gestion diffusion model from influence maximizaiton perspective.To speed up the traffic bottlenecks identification,a latent influenced crowdedness metric is designed for seeds selection.