In this work a neuro-fuzzy based model of a whey batch fermentation process by a strain Kluyveromyces marxianus var. lactis MC5 is presented. A three-layered neuro-fuzzy network is realized. The simulation results are compared with conventional models (based on mass balance and differential equations). The neuro-fuzzy model provides a better fitness and allows inclusion of linguistic variables (such as colour, smell, taste, morphophysiology, etc.). The accuracy is approximately equal to this achieved by a conventional neural network. The proposed approach is flexible (with regard to the process model) and quite robust (with regard to the possible uncertainties and to the optimization surface). Future work will focus on applying this approach for modelling of different biotechnological processes.
Keywords: Kluyveromyces marxianus var. lactis MC5; fuzzy sets theory; neural network; neuro-fuzzy neural network.