PM2.5-O3 Compound Pollution Pattern Recognition Based on Hybrid Model of tensor Decomposition and Spectral Clustering

Clustering intra-urban areas with the same pollution characteristics can help to accurately manage intra-urban air pollution. In this study, a new classification model for PM
2.5
-O
3
composite pollution features is established based on tensor decomposition and spectral clustering methods. The tensor decomposition was utilized to extract pollution features from massive air pollutant concentration data, and the pollution concentration data were decomposed into day-by-day, intra-day, and spatial modes, which provided a low-dimensional space for the subsequent spectral clustering and fine clustering classification process of the urban spatial-scale PM
2.5
-O
3
composite pollution conditions. The model is explored by taking the annual PM
2.5
and O
3
pollution conditions in Beijing in 2017 and 2020 as an example, discussing the differences in geographic locations and pollution characteristics between the different models, and explaining the changes in the city’s pollution characteristics from 2017 to 2020. Through this new hybrid method, scientific support can be provided for urban refinement and precise management.