Multi-objective optimization of structural parameters of SCR system under Eley-Rideal reaction mechanism based on machine learning coupled with response surface methodology

Abstract

Selective catalytic reduction (SCR) is an important method to control nitrogen oxides (NOx) emissions from diesel engines. Excellent SCR structural parameters are the key to effectively reduce NOx and back pressure. The dynamic reaction processes of NOx standard reaction, fast reaction and NO2-SCR reaction are deeply explored by establishing the Eley-Rideal model. The results show that the wall thickness and washcoat thickness of the SCR are the main determinants of the catalyst performance, while the CPSI has a great influence on the pressure drop. In addition, regression prediction analysis of experimental data by random forest (RF), particle swarm optimized backpropagation artificial neural network (PSOBP-ANN) and response surface methodology (RSM) was performed to explore the coupling relation functions of structural parameters, and optimal test results were solved and verified. The denitrification efficiency of the structure-optimized SCR system increased by 22 % and the pressure drop decreased by 23 %.

Keywords

• Selective catalytic reduction
• Predictive model
• Machine learning
• Multi-objective optimization