Research On Similarity Mining Of Urban Traffic Congestion Evolution

The evolution of urban congestion is an important field of research in traffic flow data mining.The scientific approach of mining congestion evolution patterns from massive historical traffic data to identify problems in traffic infrastructure and predict congestion is an important way to alleviate the traffic congestion problem in modern cities.Based on the high similarity of traffic flow data in spatial and temporal dimensions,this paper presents the research on the similarity mining of traffic congestion evolution based on the public electronic map traffic condition classification data and applies it to potential traffic bottleneck extraction and traffic congestion evolution prediction.The main research results are as follows:(1)A potential bottleneck extraction algorithm based on similarity mining of congestion evolution events is proposed.The complete process of congestion evolution from the beginning,diffusion,peak,dissipation,and disappearance of congestion in local areas is defined as a congestion evolution event,and corresponding extraction algorithms are proposed.Based on the spatial location of the congestion evolution event,the regional division of the large-scale road network was completed.By conducting spatiotemporal clustering analysis on congestion evolution events within the same region,multiple congestion evolution modes were extracted and the types of congestion evolution were divided into frequent and occasional congestion.By analyzing the evolution process of congestion,potential traffic bottlenecks that trigger this type of congestion evolution pattern were identified.(2)The traffic congestion evolution prediction algorithm based on sequence pattern mining is proposed.By considering the congestion evolution process as a sequence and taking into account the spatial topological adjacency of road segments and the strict temporality of traffic sequence data,the sequence pattern mining algorithm is proposed to take into account the spatial topological relationship and strict temporality,and a congestion evolution rule base is developed.Finally,by combining the congestion evolution rule base and the similarity of historical congestion evolution,the future traffic congestion evolution is predicted.The proposed method is validated by using the traffic condition data of the local area of the West 2nd Ring Road in Beijing from October 10 to December 10,2021,on Gaode E-map as an example.The experimental results show that: the method in this paper can extract congestion evolution events and classify congestion areas more accurately;through the spatiotemporal clustering analysis of congestion evolution events in the key study area,108 congestion evolution patterns are found and 22 potential traffic bottlenecks are extracted;based on the sequence pattern mining results and historical congestion evolution events,the future traffic conditions of the road segments involved in the congestion process are predicted and the accuracy is 86%.The method is compared with typical prediction methods such as ARIMA,LSTM and GNP-based association rule mining model,and the results prove that this method has obvious advantages.The method is important for the scientific allocation of traffic resources and for improving the efficiency of the traffic network.