Research And Application Of Clustering Algorithm For Generalized Traffic Flow

With the update of traffic information acquisition methods,traffic flow data has become more abundant,which provides a new perspective and ideas for the research of urban residents' travel laws.The extensive application of trajectory data has also injected new vitality into the research of generalized traffic flow.Faced with the diversity of urban transportation needs,the complexity of transportation networks,and the synergy of transportation behaviors,we have increasingly recognized the importance of mining the spatial and temporal patterns of traffic flow in the study of the complex transportation system of "human-vehicle-road-land use" coupling.In this paper,taking Beijing traffic performance index(TPI)and taxi origindestination(OD)flow as examples,taking the short-term prediction of macro traffic flow in the time dimension and the micro traffic flow pattern mining in the space dimension as the foothold,the application of clustering algorithm in the short-term prediction of TPI and the spatial clustering of OD flow is studied respectively.This paper proposes two clustering-based traffic flow analysis algorithms: 1)A pattern sequence forecasting algorithm(PSF)considering time decay.This algorithm recognizes traffic flow congestion patterns through clustering algorithms and implements dynamic prediction based on the time correlation and proximity relationship of traffic flow pattern sequences.In this paper,this algorithm is used for short-term forecasting of Beijing's TPI.The forecasting accuracy is better than the baseline model,and it is interpretable.2)An OD flow clustering algorithm based on vector constraints(ODFCVC).This algorithm expresses the high-dimensional characteristics of OD flow by defining OD flow event points and OD flow vectors,and uses two-step clustering to realize the pattern mining of OD data in the flow space.In this paper,the algorithm is used to recognize the OD flow pattern of taxis in Beijing.Using this algorithm,new traffic flow communities and irregular traffic flow clusters with representative flow trends are obtained.