A Physics-Informed Neural Network-Based Model for Predicting NOx Emissions of Coal-Fired Boilers

A Physics-Informed Neural Network (PINN)-based method for predicting NOx emissions in coal-fired boilers is proposed in this paper. Firstly, a data-driven neural network model and a physics-based combustion mechanism model were constructed separately. Secondly, a custom-designed loss function was introduced to embed physical constraints into the model training process, ensuring that the predicted results comply with the physical laws governing NOx emissions. Finally, the PINN was trained using actual operational data from power plants to predict NOx emission concentrations in boilers. Experimental results demonstrate that the developed PINN model significantly outperforms four benchmark models in terms of prediction accuracy, especially under conditions of fluctuating loads that induce severe changes in NOx emissions, where it exhibits superior robustness and adaptability. The proposed PINN provides a novel research approach for monitoring and predicting boiler operating data under variable working conditions, while also offering theoretical support for intelligent optimization of industrial processes.