Modeling and Optimization of NO_x Emission from a 660 MW Coal-Fired Boiler Based on the Deep Learning Algorithm

抄録

With the increasingly strict environmental protection policies, restrictions on NO_x emissions are becoming increasingly stringent. This paper focuses on modeling and optimizing NO_x emission for a coal-fired boiler with advanced deep learning approaches. Three types of deep recurrent neural network models, including recurrent neural network (RNN), long short-term memory (LSTM), and gate recurrent unit (GRU), are developed to model the relationship between operational parameters and NO_x emission of a 660 MW boiler. The hyperparameters of the models are selected by grid search and the effects of the hyperparameters on the prediction results are analyzed. Compared with the traditional back propagation neural network (BP), support vector machine (SVM) models and deep belief network (DBN), the deep recurrent neural network models have higher prediction accuracy. The experimental results show that the GRU-based NO_x prediction model has the best prediction performance among the proposed models. Then, the predicted NO_x emission is used as the objective of searching the optimal parameters for the boiler combustion through the grey wolf optimization (GWO) algorithm. The searching process of GWO is convergent. According to the simulation results, the declines in the NO_x emissions in the two selected cases were 19.49% and 17.96%, which are reasonable achievements for the boiler combustion process.