A Genetic Algorithm‐Based Method for the Automatic Reduction of Reaction Mechanisms

Abstract

An automatic method for the reduction of chemical kinetic mechanisms under specific physical or thermodynamic conditions is presented. The method relies on the genetic algorithm search logic to gradually reduce the number of reactions from the detailed mechanism while still preserving its ability to describe the overall chemistry at an acceptable error. Accuracy of the reduced mechanism is determined by comparing its solution to the solution obtained with the full mechanism under the same initial and/or physical conditions. However, not only the chemical accuracy and the size of the mechanism are considered but also the time for its solution which helps to avoid stiff and slow converging mechanisms, thus preferring the fast solutions. The reduction method is demonstrated for a detailed mechanism for methane combustion, GRI‐Mech 3.0, which was reduced from 325 reactions and 53 species to 58 reactions and 26 species, and for an iron oxide formation mechanism from iron pentacarbonyl doped flames by Wlokas et al. (Int J Chem Kinet 2013, 45(8), 487–498), originally consisting of 144 reactions and 34 species, which was reduced to 37 reactions and 24 species. The performance of the reduced mechanisms is shown for homogeneous constant pressure reactors and for burner‐stabilized flames. The results show a good agreement between reduced and full mechanisms for both the reactor and flame cases. The presented method is flexible and can be easily adjusted to either yield more accurate (but bigger) or smaller (but less accurate) reduced mechanisms, depending on the user's preference.

Analysis of Kinetic Reaction Mechanisms

“…Other approaches to solving optimisation problems in mechanism reduction include binary encoded genetic algorithms (GAs) as discussed in Edwards et al (1998), Banerjee and Ierapetritou (2003), Elliott et al (2004Elliott et al ( , 2005Elliott et al ( , 2006, Montgomery et al (2006), Hernández et al (2010) and Sikalo et al (2014). Other approaches to solving optimisation problems in mechanism reduction include binary encoded genetic algorithms (GAs) as discussed in Edwards et al (1998), Banerjee and Ierapetritou (2003), Elliott et al (2004Elliott et al ( , 2005Elliott et al ( , 2006, Montgomery et al (2006), Hernández et al (2010) and Sikalo et al (2014).…”

Optimized Reduced Chemistry and Molecular Transport for Large Eddy Simulation of Partially Premixed Combustion in a Gas Turbine

“…All these techniques have benefited from automated optimization tools. In the case of reduced chemistry, it was shown that the rates of global schemes could be found from optimization procedures (Polifke et al, 1998;Tham et al, 2008;Abou-Taouk et al, 2012;Sikalo et al, 2013;Farcy et al, 2014). In the context of look-up table construction, it was also proposed recently to determine with optimization the best fitted progress variable, based potentially on all species of a detailed chemical scheme (Niu et al, 2013;Prufert et al, 2015).…”

A Genetic Algorithm–Based Method for the Optimization of Reduced Kinetics Mechanisms

“…The optimization method was demonstrated for (further) reduced versions of the GRI‐Mech 3.0 methane combustion mechanism and a tert‐ butanol mechanism . These reduced mechanisms were obtained with a genetic algorithm–based method for a homogeneous reactor model for evaluation. It is important to note that the original GRI‐Mech 3.0 is already optimized, so that a further optimization of a mechanism derived from it might violate the uncertainty range of some reactions.…”

“…Prior to the optimization, the GRI‐Mech 3.0 mechanism had been reduced to 52 reactions and 26 species by removing 273 reactions and 27 species for a homogeneous combustion at atmospheric pressure, stoichiometric methane/air mixture, and an initial temperature of 1400 K . The reduction criteria were the ignition delay time, final temperature, number of reactions, and the computational time required for the solution of a homogeneous reactor problem . The computational time and the number of reactions were taken as the reduction criteria to avoid possible stiffness and numerical instabilities that may occur due to eliminating reactions.…”

“…The deviation between the reference ( ξ ref ) and the optimized mechanism ( ξ opt ) is normalized using the logarithmic expression (Eq. ), Figure a introduced before . …”