Deep Learning Models for SO₂ Distribution in a 30 MW Boiler via Computational Fluid Dynamics Simulation Data.

Area of Science:

• Combustion Engineering
• Chemical Engineering
• Artificial Intelligence in Engineering

Background:

• Sulfur dioxide (SO2) distribution in boilers is critical for managing furnace tube corrosion.
• Understanding the correlation between SO2 and other combustion variables is essential for process optimization.

Purpose of the Study:

• To develop a hybrid deep learning model for analyzing SO2 distribution in boilers.
• To identify key variables influencing SO2 distribution using computational fluid dynamics (CFD) data.
• To establish the relationship between SO2 and other combustion products.

Main Methods:

• Computational fluid dynamics (CFD) simulations were performed to generate training data.
• A LASSO algorithm was employed for input variable selection based on correlation with SO2 distribution.
• A deep belief network (DBN) integrated with a restricted Boltzmann machine (RBM) and a fully connected layer was utilized.

Main Results:

• The hybrid deep learning model successfully described the nonlinear relationship between SO2 distribution and other combustion variables.
• Oxygen (O2) concentration was identified as the most influential variable affecting SO2 distribution.
• This represents the first application of deep learning to correlate SO2 distribution with combustion products.

Conclusions:

• The developed deep learning model provides a novel approach to understanding SO2 distribution in boilers.
• Optimizing O2 concentration can effectively manage SO2 distribution and mitigate furnace tube corrosion.
• Further research can explore advanced deep learning architectures for combustion process analysis.