Combining machine learning to explore the influence of VOCs and meteorological factors on ozone concentration: A case study of a chemical park in Shenyang, China

Abstract

Ozone ($O_3$) pollution is a pervasive air quality issue, with volatile organic compounds (VOCs) significantly contributing to $O_3$ formation. Utilizing continuous online monitoring data from a chemical park in Shenyang in 2019, we analyzed variations in VOCs, conventional pollutants, and meteorological conditions, preliminarily determining their apparent effects on $O_3$. Results indicated an “M”-shaped monthly average ozone variation, with peaks in summer and troughs in winter, influenced by photochemical reactions and meteorological factors. The daily mean concentrations of alkanes, alkynes, and aromatic hydrocarbons in VOCs displayed clear periodicity, with peaks occurring at 5:00–7:00 and 19:00–20:00, and troughs at 14:00–15:00. The monthly mean concentrations exhibited seasonal trends, with higher levels in fall and winter, and lower levels in spring and summer, demonstrating a pattern opposite to that of $O_3$. Using machine learning techniques, we modeled the relationship between key factors and $O_3$ concentration. The results revealed that the optimized Extreme Gradient Boosting (XGBoost) model achieved the highest correlation coefficient and demonstrated the best performance. Using the feature importance method, we identified the key factors most strongly associated with $O_3$ concentration. The optimized model was then employed to examine the variation in $O_3$ concentration across different temperature and humidity conditions. This study provides essential insights for developing effective pollution control strategies and guiding environmental management decisions.