A clustering study of three-phase traffic flow based on the improved KMC algorithm

According to the operational characteristics of urban expressway traffic flow, this paper conducted a study relevant to three-phase traffic theory based on K-means clustering (KMC) algorithm. The KMC algorithm has been improved to avoid clustering into local optimum and get a better initialization of clustering center. Based on the information entropy and its transition value have been used as evaluation criteria to determine optimal classification number of the traffic flow time series. It is found that clustering results of different sections have consistency and differences through the study of the clustering of traffic flow time series of upstream and downstream related sections. The information of data based on Shannon entropy has been used for complete the validity of the classification result. The research of traffic flow time series clustering analysis in this paper give a reliable method to determine the three-phase boundary of traffic flow from the perspective of traffic flow data, and provide a theoretical and methodology support to the traffic flow operation discrimination and the traffic flow prediction.