Prediction of NOx Emissions from Coal-Fired Boilers Based on Support Vector Machines and BP Neural Networks

The BP neural network and support vector machine (SVM) are respectively employed using operation test data to establish models describing the NO_x emission characteristics of a coal-fired boiler with the assistance of the intelligent MATLAB toolbox. The momentum method is employed to improve existing problems within the BP neural network, and to choose the optimal kernel function of the SVM prediction model and the corresponding parameters c and g. The maximum error of the prediction model of the improved BP neural network is 9.85% with an average error of 4.2%; the maximum error of the SVM prediction model after parameter optimization simulation is 4.57% with an average error of 2.15%. Results indicate that both modelling methods demonstrate improved accuracy and generalization. Finally, quantitative comparison analysis of the simulation and prediction results of the two models indicate that the supporting vector machine model is greatly superior to the neural network model in terms of computing speed, fit and generalizability while requiring fewer thermal state data samples from boiler operation.