关键词: 交通流, 时间序列, 模式相似性度量

Abstract: As high dimensional and high noise of traffic flow time series, the on-line segmented algorithm which is based on changing trends, goodness of fit and minimal distance/percentage principal is designed for time series dimension reduction. On the basis of segmentation results, the time-series is represented as 5 tuples piecewise linear form. Then five shape-
similarity distances of time-series are defined, and whose performance on discriminate of different traffic flow pattern are tested through cohesion hierarchical clustering. Taking the loop detector data from the eastern of Shanghai North-South expressway as an example, the method is verified. The combination of pattern distance and Euclidean distance is determined as the best method for pattern similarity measurement of traffic flow time series.

Key words: traffic flow data, time series, pattern similarity measurement