Machine Learning-Assisted Prediction and Exploration of the Homogeneous Oxidation of Mercury in Coal Combustion Flue Gas

Mercury emission from coal combustion flue gas is a significant
environmental concern due to its detrimental effects on ecosystems
and human health. Elemental mercury (Hg0) is the dominant
species in flue gas and is hard to immobilize. Therefore, it is necessary
to comprehend the reaction mechanisms of Hg0 oxidation,
namely, Hg0 to oxidized mercury (Hg2+), for
mercury immobilization. In spite of extensive research on homogeneous
Hg0 oxidation, universal accurate prediction models and
unified explanations are lacking. In this study, for the first time,
quantitative prediction models were developed for the Hg0 oxidation percentage with machine learning (ML) using flue gas compositions
and operating conditions as inputs. Gradient boosting regression models
showed optimal performance (test R2 ≥
0.85). ML-aided feature analysis results exhibited that Cl2, HCl, Hg0, temperature, and HBr were the top five critical
factors affecting mercury homogeneous oxidation. Halogen gas promoted
Hg0 oxidation at temperatures around 250 °C, while
Hg0, SO2, and quench rates were not conducive
to Hg0 oxidation. High reaction rate coefficients for the
Hg/Cl and Hg/Br reactions verified the ML interpretive results and
revealed the major mercury homogeneous oxidation mechanisms. Models
developed here may play important roles in understanding Hg0 oxidation and optimizing flue gas Hg immobilization technologies.