Sequence clustering analysis of air pollution based on adaptive radius determination and automatic density peaks detection

DPClust is a new clustering algorithm with the characteristics of fast search and based on local density of data point. However, the parameter radius was determined by experience and the cluster centers were selected in a manual way. The intention of this study is to combine both adaptive radius threshold determination and automatic density peak detection together with DPClust. This paper proposes an improved method which combines data field theory and linear classification to calculate radius adaptively and select cluster centers automatically. The proposed method has been validated by analyzing PM2.5 sequence data of 362 cities in China. The method can complete the clustering process in one step instead of iteration. The sequence clustering method proposed in this paper can also applied in other air pollutant sequence to mine more regular patterns of air pollution in large-scale monitoring points.