关键词: 共享单车, 时空大数据, 细粒度聚类, DBSCAN, K-means

Abstract: Aiming at the problem that the clustering results obtained by the traditional methods such as K-means and DBSCAN alone when clustering the location data of shared bikes are inconsistent with the real clustering structure, a fine-grained clustering method(FGCM) based on spatio-temporal big data of shared bikes is proposed. This method uses DBSCAN to perform initial clustering, and on this basis, uses GMM-EM algorithm to perform detailed clustering to extract fine-grained hotspots. Experiments show that this method can eliminate noise and outliers based on the density threshold, has no need to specify the number of detailed clusters, and the shape and size of the clusters are more flexible. In the case of clustering the location features of the big data of shared bikes, compared with traditional methods that use K-means or DBSCAN alone, FGCM has a higher degree of refinement, and can fully demonstrate the actual characteristics of shared bikes, which can be used to plan facilities such as electronic fences and helps to regulate the parking of shared bikes without reducing the commuting efficiency.

Key words: shared bikes, spatio-temporal big data, fine-grained clustering, DBSCAN, K-means