Research on digital twin modeling method for combustion process based on model reduction

In response to the difficulty in obtaining combustion information within coal-fired boiler furnaces, a method is proposed in this study to improve the reduced-order model using clustering segmentation. This approach aims to rapidly predict the combustion temperature field inside the furnace by establishing a twin model of the combustion temperature field. Initially, the finite volume method is employed to analyze the combustion system of a 600MW subcritical boiler under various operating conditions. Subsequently, cross-sectional data from burner nozzle positions at each operating condition are extracted. These data are subjected to Proper Orthogonal Decomposition (POD), Spectral Proper Orthogonal Decomposition (SPOD), and Wavelet Transform-POD (WT-POD) for dimensionality reduction to obtain modal data. Comparative analyses are conducted on the modal data obtained from different methods. Furthermore, based on modal data analysis, a Support Vector Machine (SVM) regression model is selected to reconstruct the temperature field. The average absolute error of the reconstructed temperature fields from three methods under different operating conditions is then compared. Finally, the model is refined using clustering segmentation, resulting in an improvement of approximately 0.6% in reconstruction accuracy. This enhancement demonstrates that the clustered POD-SVR-GA model achieves higher accuracy in reconstructing combustion temperature fields after clustering-based improvements.