A hybrid genetic algorithm with local search - II. Continuous variables: multibatch peak deconvolution

A hybrid genetic algorithm with internal local search was developed for opt imisations involving continuous variables. The reproduction probabilities w ere enhanced using the fitness values obtained when a local method was appl ied to each individual in the population. These estimations are more realis tic, since consider not the apparent but the hidden, latent quality of each individual. The information gathered in the local search was also used to build an auxiliary population recording the successfully enhanced individua ls, which allowed to detect the convergence and self-adapt the search limit s. The size of this auxiliary population was kept constant by a cluster ana lysis strategy. The method was applied to the simultaneous deconvolution of sets of chromatograms monitored at a single detection wavelength, sharing two compounds (sulphapyridine and sulphisoxazole) at different concentratio n ratios. The results were compared with a classical genetic algorithm and a hybrid Powell-Gauss-Newton method, to check the benefits of the strategy. The method, called LOGA (locally optimised genetic algorithm) was superior in terms of the obtained residuals, and was able to retrieve the expected individual peak profiles with very low errors. (C) 2001 Elsevier Science B. V. All rights reserved.