Fluctuation Similarity Modeling for Traffic Flow Time Series: A Clustering Approach

Traffic time series analysis is important because of its use in traffic control and travel time prediction. In this paper, we discuss how to cluster traffic time series that have similar fluctuation patterns. We use simple average detrending method and only study the residual time series. Second, we use principle component analysis (PCA) on raw data and use the weight of the first d-components as the features of the time series. Third, we use k-means algorithm to cluster the traffic time series. Finally, we study the results of the clustering algorithm and discuss the origins of the clusters. In summary, the most important factors of clustering results are urban/rural area, direction and in/not in ramp entrance.