文献类型：期刊文献

英文题名：Reaction kinetics-guided multi-mode feature fusion Transformer for pollutant concentration prediction

作者：Huang, Hengling[1];Deng, Kai[3];Wang, Luyao[2];Guo, Wenze[2];Long, Jian[2]

机构：[1]East China Univ Sci & Technol, Sch Chem Engn, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai, Peoples R China;[3]Shanghai Marine Diesel Engine Res Inst, Shanghai 201108, Peoples R China

年份：2026

卷号：14

期号：5

外文期刊名：JOURNAL OF ENVIRONMENTAL CHEMICAL ENGINEERING

收录：;EI(收录号：20262320866479);WOS:【SCI-EXPANDED(收录号:WOS:001793027400001)】；

基金：Acknowledgments This work was supported by the National Natural Science Foundation of China (62373155, 22408099) .

语种：英文

外文关键词：Feature fusion Transformer; Attention mechanism; Reactor modeling; Hybrid modeling; Denitrification; Desulfurization

摘要：Emissions of NOX and SO2 from fluid catalytic cracking (FCC) regenerators pose significant environmental and health risks, making accurate predictions essential for sustainable and resource-efficient flue gas treatment. The denitrification (DeNOX) and desulfurization (DeSO2) processes exhibit strong nonlinearity due to multiphase flow, complex reaction networks, and heat transfer interactions, which challenge conventional modeling approaches. To address these issues, a hybrid framework (KTIHybrid) integrating reaction kinetics with a multi-mode feature fusion Transformer is proposed for outlet concentration prediction. Prior to model construction, industrial data are systematically preprocessed to enhance robustness and capture multi-mode operating characteristics. Subsequently, a mechanistic model based on reaction kinetics (RKMech) is established and optimized, and its predicted outputs are introduced as auxiliary inputs to the multi-mode feature fusion Transformer (MFFTrans). Validation using real plant data demonstrates high predictive accuracy for both DeNOX (MSE = 0.92, R2 = 0.98) and DeSO2 (MSE = 0.009, R2 = 0.998) processes. The enhanced prediction stability supports optimized reagent utilization and sustainable emission control, contributing to low-emission and energy-efficient refinery operations.