(571g) Artificial Neural Networks for Flare Modeling and Set Point Determination | AIChE

(571g) Artificial Neural Networks for Flare Modeling and Set Point Determination


Damodara, V. - Presenter, Lamar University
Chen, D., Lamar University
Lou, H. H., Lamar University
Alphones, A., Lamar University
Martin, C. B., Lamar University
Li, X., Lamar University
The refineries and chemical industries are mandated to meet the EPA’s regulations for flare performance, i.e., smokeless flaring while maintaining a minimum combustion zone heating value of 270 BTU/scf and NHVdil >22 Btu/ft^2 for all steam/air/non-assisted flares. While field tests of the many flare operation parameters are expensive, it is highly efficient to use mathematical models to analyse data and develop the right tools to make decisions in the real plant scenario. Modelling flare data using artificial neural networks to achieve the desired combustion efficiency (CE) and eliminate smoke can be done robustly. Selected flare data obtained from various flare tests having a wide range of exit velocity, heating value, and fuel composition have been analysed and the most influencing independent variables are chosen. Both steam and air assisted flares were modelled using the neural network toolbox in MATLAB with high correlation coefficients of greater than 0.95. Determining the set point (amount of steam/air/make-up fuel required) at the Incipient Smoke Point (ISP) and for smokeless flaring has been performed as a part of this study. Desirable operating inputs can be set for the ISP (NHVdil >22 BTU/ft2  & Opacity<Opacity ISP) and for smokeless flaring with a high CE (>96.5).