Prediction of NOx emission for coal-fired boilers based on deep belief network

This study developed three types of deep belief network (DBN)-based models to estimate NOx emission in coal-fired power plants by a new data acquisition method. Based on the experimental data obtained by field experiments, the validated Fluent-based simulation results and the historical operating data from a database are used in the model calculations. Using mutual information, the model input set is optimized by the feature selection method. With the optimal inputs, three types of DBN-based NOx prediction models are constructed, in which the extreme learning machine, back propagation network, and radial basis function network are below the top layer of the DBN to serve as the regression model. The results indicate that the DBN-based models have a greater prediction accuracy with 0.93,0.9, and 0.89 coefficients of determination and greater robustness compared to the three other NOx prediction models.