Modifications of the classical genetic algorithm for non-linear fitting applied to response surface modeling in HPLC Original Research Article

Two modifications of the classical genetic algorithm for non-linear fitting are presented and discussed. The first attempts a unification of Brunettiʹs algorithm with the classical genetic algorithm and its basic feature is that mutations with a certain probability are created only when the variance of the cost values of the population goes below a preset value. The second modification is referred to systems that exhibit more than one local minima (maxima) and the convergence of the classical genetic algorithm to these minima (maxima) is rapid enough. Then, the performance of the classical algorithm may be enhanced by creating new populations when the number of generations gets certain values. Using HPLC retention data we found that the performance of our modifications is better than that of previous genetic algorithms.