Statistical and Neural Network Modeling of FLARE Performance and Estimation of Controllable Parameters

Recent EPA proposed rules (parts of 40CFR63.11 & 60.18) apply maximum allowable flare tip velocity (400 ft/s), smokeless flaring, and combustion zone heating value at a minimum value of 270 BTU/scf for all steam assisted, air assisted and non-assisted flares including emergency release to achieve high combustion and destruction efficiency (CE/DE). The proposal was based on analysis conducted on a limited set of data from steam assisted and air assisted flares. In this study, the flare soot emission, visible emission scale, DE, and CE test data from 1983 to 2014 (including 1983/1984 EPA, 2010 TCEQ, 2009/2010 Marathon Detroit/Texas City, and 2014 Carleton University) were analyzed. This study aims at developing nonlinear statistical models for flare efficiencies and opacity. All CE data were corrected for soot emissions. Curve fitting and Neural Network tool boxes in MATLAB were used in modeling. Multivariate statistical distribution functions and neural network models were developed for steam and air assisted flares burning propylene, propane, natural gas, methane and ethylene. Since the flare test data includes larger range of exit velocity which affects plume volume, opacity is normalized with plume diameter. The input variables include vent gas combustibility (combustion zone net heating value (NHVcz)/ lower flammability limit (LFLcz)/ combustible concentration (Ccz)), vent gas species (olefins, alkynes, aromatics, hydrogen, nitrogen, carbon to hydrogen ratio), and operating parameters (steam/air assists, and exit velocity). 3-D contour plots were developed to estimate the desirable range of controllable operating variables. The goal is to estimate the desirable operating set point with a high CE >96.5% (or DE >98%) and low opacity (10% for steam assist & 40% for air assist). Neural network models developed for CE and normalized opacity showed high coefficient of determination (R²) values 0.91 and 0.95 respectively for steam assisted flare data.

 

Keywords: Propylene, Propane, Methane, Ethylene, Natural gas, visible emissions, Flare performance, Lower flammability limit, Net heating value at combustion zone, Neural network models.