Efficient Clustering-Based Genetic Algorithms in Chemical Kinetic Modelling

Two efficient clustering-based genetic algorithms are developed for the optimisation of reaction rate parameters in chemical kinetic modelling. The genetic algorithms employed are used to determine new reaction rate coefficients for the combustion of four different fuel/air mixtures in a perfectly stirred reactor (PSR). The incorporation of clustering into the genetic algorithm allows for a considerable reduction in the number of computationally expensive fitness evaluations to be realised without any loss in performance. At each generation, the individuals are clustered into several groups and then only the individual that represents the cluster is evaluated using the expensive fitness function. The fitness values of the other individuals in the same cluster are estimated from the representative individual based on a distance measure in a process called fitness imitation.