Dominant Factor Identification and Predictive Modeling of PM2.5-Bound Sulfate from Chinese Coal-Fired Power Plants

Abstract

PM_2.5-bound sulfate (p-SO₄^2–) from coal-fired power plants (CFPPs) is a crucial component of atmospheric particulate matter, and its formation is comprehensively influenced by coal composition characteristics and air pollution control devices (APCDs). Based on a data set containing 109 measured mass fractions of p-SO₄^2– (p-SO₄^2– fraction) from CFPPs in China, this study develops a Bayesian linear regression model to identify the dominant factors of p-SO₄^2– formation and to quantify the effects. The results indicate that coal’s sulfur content and usage of certain APCDs promote the formation of p-SO₄^2–, including selective catalytic reduction (SCR), wet flue gas desulfurization (WFGD), and semidry desulfurization (SDD), whereas the wet electrostatic precipitator (WESP) and desulfurization efficiency inhibit it. Benchmarking against machine learning approaches demonstrates the performance of the Bayesian model (R² = 0.72, and R_LOO² = 0.45), which outperformed random forest and XGBoost algorithms in generalization ability, showing its advantages in addressing small data sets. The model predicts an average p-SO₄^2– fraction of 0.144 ± 0.037 g/g across 69 CFPPs in the Beijing–Tianjin–Hebei (BTH) region. This study systematically evaluated the roles of multiple influencing factors on p-SO₄^2– formation and predicted the p-SO₄^2– fractions derived from CFPPs in the BTH region, providing a quantitative decision-making basis for precise sulfate emission control in CFPPs and regional environmental planning.