The paper proposes a new method for transforming traffic flow time series into the vector symbolization series based on data mining technique. With consideration of the complexity and randomness of traffic system, the linearization smooth of discrete traffic flows is developed to filter stochastic flows by least square approximation, and the extensive meaning regularity and knowledge of traffic data are drawn. It segments the time series of traffic flows into linear segments that are well applied for their shape expression. Then online clustering analysis on the segmented traffic flow time series is conducted step by step with the similarity threshold and the improved K-Means algorithm to adapt the characteristics of traffic data. The measure of shape similarity is also proposed in this paper. Effectiveness of this method has been verified by the vector symbolization series of express highway traffic flows. The influencing factors of experimental result are discussed, some research to be taken in the future is proposed.
Research on vector symbolization of traffic flow time series by data mining oriented method
2003-01-01
364831 byte
Conference paper
Electronic Resource
English
Research on Vector Symbolization of Traffic Flow Time Series by Data Mining Oriented Method
British Library Conference Proceedings | 2003
|Traffic Flow Management: Data Mining Update
NTIS | 2012
|Traffic Flow Management: Data Mining Update
NTRS | 2012
|