Abstract: Congestion road segments over real GPS trajectory data was predicted. The traditional methods were ostracized based on traffic flow prediction and congestion identification， and a novel method was proposed based on the congestion vector and the congestion transition matrix. The congestion vector and the congestion transition matrix were established by mining and training taxi GPS trajectory data， resulting in the implementation of the prediction of traffic congestion. In the course of prediction， time periodicity and spatial-temporal correlation of road segment congestion were both considered. The experiments on real data show the effectiveness of the congestion road segment prediction method proposed.

Key words: GPS trajectory data, spatial-temporal causal relationship between road segments, congestion transition probability, congestion transition matrix, congestion road segment prediction