Abstract:

The occupancy data of a small region contains 29 loops in Beijing city was analyzed using generative topographic mapping through time(GTM-TT). The original data was mapped from 29 dimensional space to 2 dimensional space which is easy for visualization and clustered into different classes naturally. The statistics of states transfer show that the most probably transfer distance was within 1 step. At last the 89 days was clustered into 4 types using hierarchical clustering method according to the Euclidean distance between the tracks of different days in the latent space. It was found that the day of different type had different traffic process.

Key words: engineering of communication and transportation, multivariate time series, GTM-TT, traffic status analysis, unsupervised clustering, data visualization