Integrated DNN and CFD model for real-time prediction of furnace waterwall slagging of coal-fired boiler

Abstract

Slagging of furnace waterwall adversely affects the thermal performance and increases the operation and maintenance costs of coal-fired boilers. A real-time slagging prediction model is of great significance for the plant operators to monitor the furnace slagging conditions and make appropriate operation of soot blowers to maintain the cleanliness of waterwall. Three-dimensional (3D) computational fluid dynamics (CFD) can predict detailed furnace slagging distributions, but its high computational cost has been the major impediment hindering its application in large-scale combustion systems that require real-time information. To resolve this problem, a CFD-based deep neural network (DNN) model framework was developed in this study to realize the real-time prediction of furnace wall slagging distribution. A 3D CFD model was first developed and employed to generate the slagging database under typical boiler operating parameters. Then this database was utilized to train the DNN model to capture the mapping relationship between boiler operating parameters and wall slagging distributions. The core idea of this CFD-based DNN model is to generalize the wall slagging distributions under different boiler operating conditions based on this mapping relationship learned by the DNN model. This model was applied to a 350 MW coal-fired boiler. The prediction results by the CFD model demonstrated that the boiler operating parameters, such as boiler load and coal burner swirling blade angle, have strong effects on furnace wall slagging distributions, while the DNN model could precisely capture their relationship and make predictions with deviation from the CFD results below 5 %. More importantly, the DNN model could give the prediction results within seconds that is three to four orders of magnitude faster than the CFD model. Thus, it can respond to the rapidly changing boiler operating conditions and realize the real-time prediction of furnace wall slagging distributions.