Fast prediction and optimization of oxy-fuel combustion in a MW-scale circulating fluidized bed: an integrated CFD and machine learning approach

Oxy-fuel circulating fluidized bed (CFB) technology demonstrates significant potential for efficient carbon capture; however, its pilot-scale investigations entail substantial costs. To reduce time and costs associated with pilot-scale testing, this study presents an efficient methodology that integrates numerical simulation with machine learning for fast prediction and parameter optimization. First, a computational fluid dynamics (CFD) model for a MWth-scale CFB oxy-fuel combustor was developed based on the multiphase particle-in-cell (MP-PIC) method. The model was validated using experimental data from the MWth-scale facility, and the simulation results elucidated the distribution and evolution of key parameters governing gas-solid flow, heat transfer, and reaction processes in the furnace. Then, by integrating numerical simulation and design of experiment, a database of 120 operating conditions of oxy-fuel CFB combustion characteristics was constructed for training and testing of a neural network model. The key parameters predicted by the neural network are in good agreement with those predicted by CFD model. Correlation analysis of variables indicates that the biomass blending ratio is negatively correlated with furnace temperature and CO2 concentration, while positively correlated with NO emissions. Finally, a genetic algorithm of multi-objective optimization was performed to maximize CO2 concentration and minimize pollutant emissions, resulting in a set of optimal solutions. The combined CFD simulation and machine learning method developed in this study aids engineers in fast assessment and optimization of oxy-fuel CFB combustion.