Comparison of Artificial Neuro Network, Least Squares Support Vector Machine and Partial Least Squares Modelling on NOx Emission

This paper deals with modelling on the nitrogen oxides (NOx) emission of a 600MW coal-fired boiler using artificial neural network (ANN), least squares support vector machine (LSSVM) and partial least squares (PLS) methods based on the experimental data. Some comparisons on the prediction accuracy, time consuming and some other aspects are also given. Two simulation cases are investigated to make a convincing comparison. One operation point is used to test the model in the first case and the other case is carried out to forecast two operation points. The results show that the ANN gives the best training accuracy; LSSVM exhibits a moderate activity and PLS is least time-consuming and easy to explain the importance of the independent variables.