Analysis Of Traffic Congestion Evolution Process Based On Temporal And Spatial Causality Mining

With the rapid growth of motor vehicle ownership in cities,traffic congestion is increasingly affecting people’s lives and urban development.How to identify and control congestion has become one of the mainstream research directions in the field of transportation.The existing traffic congestion control theory lacks the evolution process tracing of traffic congestion.In this way,it is impossible to fully grasp the dynamic balance relationship of the traffic flow in the controlled area.In view of the above problems,this paper proposes a method for tracing the origin and historical propagation path of congestion,and predicting the future evolution path.The method uses the transfer entropy to characterize the causal relationship between the states of the road segments in the road network,to reflect the dynamic transmission process of congestion between road sections.Specifically,the research content mainly includes four points:(1)The process and causes of urban traffic congestion are analyzed in a full cycle.From the perspective of urban land use and demand occurrence,urban management,supply and demand balance,changes in road conditions,the causes of congestion are analyzed,and further from the point congestion,line congestion and area congestion evolution path,according to different types of congestion,the whole process of congestion from formation,spread to dissipation is identified,and the process is described by the concept of congestion dynamic cause and effect tree.(2)In order to identify the road sections and time periods that congestion frequently occurs in the urban road network,a spatiotemporal pattern clustering algorithm using various traffic parameters is proposed.Firstly,a spatiotemporal regularization matrix factorization method is proposed,which can realize the completion of missing data,and can also extract the temporal feature matrix and the spatial feature matrix from the perspective of space and time.The algorithm introduces the ARIMA model in the time dimension as a space-time regular term.Secondly,the paper proposes the tensor t-SNE algorithm,based on temporal feature tensor and spatial feature tensor to further extract spatiotemporal patterns.This algorithm can solve the fusion problem of different feature parameters and the overredundancy problem of state feature parameters.Its advantage lies in that the convergence semantic features can be retained and mutually exclusive semantics can be retained and the mutually exclusive semantics can be filtered while the fusion features,achieving a more appropriate fusion of different parameters.(3)A congestion history propagation path tracing algorithm based on spatiotemporal causality search is proposed.The transfer entropy theory is introduced to characterize the causality between variables,and the breadth-first network search algorithm is introduced.Combining the breadth-first network search algorithm with transfer entropy,a spatial-temporal congestion causality tracing algorithm based on transfer entropy is proposed,which adapt to the changing characteristic of congestion.Finally,the actual application effect of the algorithm is verified by using the GPS data of Didi.com car-hailing in Chengdu.(4)In order to further clarify the development and changes of congestion in the future,combined with the multi-task machine learning algorithm,a prediction algorithm for the future propagation path of congestion is proposed.Combined with the congestion tracing model,it can form an understanding of the whole process of congestion evolution.In this method,the speed and transfer entropy are jointly predicted by using transfer entropy which can link the speed of different road sections in the past and the future,and a multi-task learning framework is proposed.On the one hand,the machine learning framework uses transfer entropy to represent spatiotemporal causality to assist in the prediction of speed;on the other hand,it uses transfer entropy formula to improve the accuracy of transfer entropy prediction.Finally,according to the prediction result of velocity-transfer entropy,the future propagation path of congestion is obtained by causality search.