Comparison of GA and PSO performance in parameter estimation of microbial growth models: a case-study using experimental data

In this work we examined the performance of two evolutionary algorithms, a genetic algorithm (GA) and particle swarm optimization (PSO), in the estimation of the parameters of a model for the growth kinetics of the yeast Debaryomyces hansenii. Fitting the model’s predictions simultaneously to three replicates of the same experiment, we used the variability among replicates as a criterion to evaluate the optimization result. The performance of the two algorithms was tested using 12 distinct settings for their operating parameters and running each of them 20 times. For the GA, the crossover fraction, crossover function and magnitude of mutation throughout the run of the algorithm were tested; for the PSO, we tested swarms with 3 different types of convergence behavior - convergent with and without oscillations and divergent - and also varied the relative weights of the local and global acceleration constants. The best objective function values were obtained when the PSO fell in the zone of convergence with oscillations or zigzagging, and had a local acceleration larger than the global acceleration. immunization