On Clustering Methods Of Traffic Flow Time Series And Applications

With the development of traffic industry, there are huge amounts of traffic flow data storing in the traffic systems everyday. It will have great significance for guiding the traffic planning and solving traffic problems, such as traffic congestion, by processing and analyzing these data properly to find valuable traffic rules. Clustering of time series, which comes from data mining technology, is an effective approach to study and analyse the traffic flow data.In this thesis, we clustered the traffic data as time series in order to find valuable traffic rules. Taking Shenzhen traffic system as the object of study, the main work is analysising the traffic flow data using clustering methods of time series. First of all, we put the traffic flow data of the whole city traffic system under pretreatment and compared the different application effect by using different kinds of typical clustering methods to cluster the traffic data of three freeways-Shennan Avenue, Binhai Avenue, and Binhe Avenue. Analysing the traffic flow data of many monitoring points on the freeways using hierarchical clustering, we found that the flow was different on different freeways. Secondly, we analysed the information of early peak and late peak of many monitoring points on freeways and divided them into different mode by using k-means clustering. Depending on the directions of traffic, the monitoring points were divided into the Early-peak mode and Late-peak mode. Finally, we compared the similarity of traffic flow time series of different monitoring points on highway and freeway, respectively, and found the different traffic characteristics between them. Compared with freeway, the traffic flow almost kept a certain level between adjacent monitoring points everyday but the similarity of different days’traffic flow of the same monitoring point on the highway was different.