Multipopulation Genetic Algorithms: A Tool for Parameter Optimization of Cultivation Processes Models

Abstract

This paper endeavors to show that genetic algorithms, namely Multipopulation genetic algorithms (MpGA), are of great utility in cases where complex cultivation process models have to be identified and, therefore, rational choices have to be made. A system of five ordinary differential equations is proposed to model biomass growth, glucose utilization and acetate formation. Parameter optimization is carried out using experimental data set from an E. coli cultivation. Several conventional algorithms for parameter identification (Gauss-Newton, Simplex Search and Steepest Descent) are compared to the MpGA. A general comment on this study is that traditional optimization methods are generally not universal and the most successful optimization algorithms on any particular domain, especially for the parameter optimization considered here. They have been fairly successful at solving problems of type which exhibit bad behavior like multimodal or nondifferentiable for more conventional based techniques.