Parallel genetic algorithms for optimised fuzzy modelling with application to a fermentation process

This paper reports the construction and application of an evolution program to a computational intelligence system used as a software 'sensor' in state-estimation and prediction of biomass concentration in a fermentation process. A fuzzy logic system (FLS) is used as a computational engine to 'infer' the production of biomass from variables easily measured on-line. For this purpose, genetic algorithms (GAs) are employed to train and tune the desired parameters of the fuzzy logic system. It is shown that the fuzzy logic system, which was tuned by two genetic algorithms implemented in parallel, produces better results in prediction of biomass concentration. The mean sum of squared errors and graphical fit are used to compare the performance of the genetically optimised FLS with artificial neural networks (ANN), which is trained using Levenberg-Marquardt second-order nonlinear optimisation method.