Computational intelligence approach for NOx emissions minimization in a 30 MW premixed gas burner.

Artificial intelligence algorithms have become a research hotspot in attempts to reduce NOx emissions in gas burners through NOx emission modeling and optimizing operating pa- rameters. This paper compres the predictive accuracy of NOx emission models based on LSSVM, SVR and ELM. CGA and three other GA based hybrid algorithms proposed to modify CGA were employed to optimize the operating parameters of a 30MW gas burner in order to reduce NOx emission. The results show that the NOx emission model built by LSSVM is more accurate than that of SVR and ELM. The mean relative error and correlation coeficient obtained by the LSSVM model were 0.0731% and 0.999, respectively. Among the four optimization algorithms, the novel TSGA proposed in this paper showed its superiority over the other three algorithms, excelling in its global searching ability and stability. The LSSVM plus TSGA method is a potential combination for predicting and reducing NOx emission by optimizing the operating parameters for the gas burner on-line. [ABSTRACT FROM AUTHOR]