Machine learning-aided optimization of coal decoupling combustion for lowering NO and CO emissions simultaneously

Abstract

Decoupling combustion technology enables significant suppression of NOx and CO emissions from solid fuel combustion, but calls for optimizing reactor structure to make full use of its superiority. Taking a coal stove as an example, three different network models were established and trained to predict the steady-state NO and CO emissions from coal decoupling combustion well. The two GRU-DNN models have higher prediction accuracy and better generalization ability than the DNN model, but they both need to be fed with complex sequence data, leading to long training and response time to new inputs. The DNN model with simple fuel properties and structural parameters as the inputs was used to forecast the steady-state NO and CO emissions from various coal-stove combinations with acceptable accuracy, so facilitating the optimization of stove structure and further coal decoupling combustion to lower the NO and CO emissions simultaneously.