Coupled simulation and deep-learning prediction of combustion and heat transfer processes in supercritical CO2 CFB boiler.

[Display omitted] • The furnace-side combustion process is simulated coupled with S-CO 2 heat transfer. • A new method to predict simulation results via RBF neural network is proposed. • 3D Eulerian-Lagrangian model computationally facilitated by the MP-PIC scheme. • Heat transfer characteristics of the novel S-CO 2 CFB boundary are studied. • Predicted and simulated results are compared to verify the neural network accuracy. The supercritical CO 2 (S-CO 2) power cycle has a wide application prospect in coal-fired power generation field because it's highly effective, compactly structured, and flexible of operation. To observe more accurate heat transfer and coal combustion characteristics in the circulating fluidized bed (CFB) with the distinctive S-CO 2 boundary, a 3D computational fluid dynamics (CFD) simulation of the furnace-side combustion process treated by the multiphase particle-in-cell (MP-PIC) method was conducted in a 600 MW S-CO 2 CFB boiler coupled with the heat transfer process on working fluid side based on the polynomial fitting calculation model. Furthermore, a novel method to predict simulation results via Radial Base Function (RBF) neural network was proposed to simplify the computational process, enhance the sample data fusion, and improve the prediction accuracy. Results show that staggered high-temperature fluid and high heat flux was a major concern in S-CO 2 heating surface arrangement. The temperature rise of wall heaters was less than the conventional steam CFB, and the heat flux of spiral and vertical heat transfer tubes decreased along the tube. The predicted temperature distribution of tubes and cold walls was in a good agreement with the coupling simulation results, whose accuracy can meet the engineering requirements. [ABSTRACT FROM AUTHOR]