Fast prediction of combustion power field in 600 MW coal-fired boiler based on CFD -AI

In this work a 600MW opposed wall-fired boiler was chosen and the numerical simulation results of CFD and artificial intelligence technology were integrated to construct a fast prediction method. The Proper Orthogonal Decomposition (POD) method was chosen to downscale the combustion power field data of the coal-fired boiler. Five different deep learning models were used to enable fast prediction of combustion power fields. The results show that the BiLSTM-Attention model had the highest prediction accuracy and shortest computation time for the temperature and heat flux density field with a relative root square error (RRSE) below 1.8% and 1.73%, respectively. The computational time for the predictive model is within 165 seconds, equivalent to 1/1025 of the time required by the CFD. The effects of the primary and secondary airflows on the wall temperature and heat flux density were also analyzed. This graphical abstract illustrates our fast prediction of the combustion power field framework for a coal-fired boiler. The framework integrates the POD method for combustion field compression. Five different deep learning models (BP, SVM, GRP, RNN, and BiLSTM-Attention models) were used to enable fast prediction of combustion power fields. The SVD-POD-BiLSTM-Attention model predicts the temperature and heat flow density field with a relative root square error (RRSE) below 1.8% and 1.73%, respectively.