Abstract:

A simple spatiotemporal identification method for PM_2.5 pollution control based on a machine learning model is proposed. A spatiotemporal clustering algorithm was used to cluster PM_2.5 in the winter prevention period, and different PM_2.5 polluted areas were identified. Furthermore, a decision tree model was constructed using the meteorological data of different areas to identify the most unfavorable influences on PM_2.5 concentration in different areas. The changes in PM_2.5 under the most unfavorable meteorological conditions were analyzed, and the optimal time period for PM_2.5 pollution control in the different study areas was determined. Using the Guanzhong area as an example, the correlation analysis of the spatial clustering of mean daily PM_2.5 in winter showed that the Guanzhong area is mainly divided into a low-altitude plain area (the Guanzhong Plain) and a relatively high mountain range. The classification tree model was then constructed, using the meteorological data of the plain and mountain areas. The analysis showed that the mountain elevations have a higher PM_2.5 concentrations under Ⅰ-10 (1.57 h ≤ sunshine hours < 7.88 h; maximum wind speed < 3.72 m·s^-1) and Ⅰ-11 (sunshine hours < 1.57 h; maximum wind speed < 3.72 m·s^-1) meteorological conditions; low-altitude areas have a higher PM_2.5 concentrations under Ⅱ-10 (small-scale evaporation ≥ 0.96 mm; average relative humidity ≥ 45.38%; sunshine hours < 8.55 h; average wind speed ≥ 2.43 m·s^-1) and Ⅱ-11 (small evaporation < 0.96 mm) meteorological conditions. Lastly, the regression analysis results showed that in the mountain elevations area in the Ⅰ-10, Ⅰ-11 category and the plain altitude area under the most unfavorable meteorological conditions (Ⅱ-10 and Ⅱ-11 category), the PM_2.5 concentration will continue to increase for 4.76 days on average, before reaching a maximum concentration.

Key words: decision tree model, PM_2.5, meteorological classification, heavy pollution