A Genetic Algorithm-Based Method for the Optimization of Reduced Kinetics Mechanisms
This paper describes an automatic method for the optimization of reaction rate constants of reduced reaction mechanisms. The optimization technique is based on a genetic algorithm that aims at finding new reaction rate coefficients that minimize the error introduced by the preceding reduction process. The error is defined by an objective function that covers regions of interest where the reduced mechanism may deviate from the original mechanism. The mechanism's performance is assessed for homogeneous reactor or laminar-flame simulations against the results obtained from a given reference—the original mechanism, another detailed mechanism, or experimental data, if available. The overall objective function directs the search towards more accurate reduced mechanisms that are valid for a given set of operating conditions. An optional feature to the objective function is a penalty term that permits to minimize the change to the reaction coefficients, keeping them as close as possible to the original value. This means that the penalty function can be used to constrain the reaction rates modifications during the optimization if needed. It is demonstrated that the penalty function is successful and can be combined with predefined uncertainty bounds for each reaction of the mechanism. In addition, the penalty function can be modified to achieve a further reduction of the mechanism. The algorithm is demonstrated for the optimization of a previously reduced variant of the GRI-Mech 3.0, a tert-butanol combustion mechanism by Sarathy et al. (Combust. Flame, 2012, 159, 2028–2055) and a hydrogen mechanism by Konnov (Combust. Flame, 2008, 152, 507–528), for which the complete uncertainty vector is known. The method has shown to be, robust, flexible, and suitable for a wide range of operating conditions by using multiple criteria simultaneously.