Modelling and optimization of biotechnological processes

En savoir plus sur le livre

Most industrial biotechnological processes are operated empirically, facing challenges due to their highly nonlinear nature. This work explores artificial intelligence methods, particularly genetic algorithms and neural networks, for the monitoring, modeling, and optimization of fed-batch fermentation processes. The primary goal is to maximize the final product while minimizing development and production costs. The interdisciplinary approach integrates biotechnology, artificial intelligence, system identification, process monitoring, modeling, and optimal control. Both simulation and experimental validation demonstrate the proposed methodologies' effectiveness. An online biomass sensor, developed using a current neural network, predicts biomass concentration with only three measurements (dissolved oxygen, volume, and feed rate). Results indicate that this sensor performs comparably or better than others requiring more measurements. Biotechnological processes are modeled using two cascading recurrent neural networks, achieving high accuracy. Optimization of the final product is accomplished through modified genetic algorithms to identify optimal feed rate profiles. Experimental production yields confirm that genetic algorithms are effective for optimizing highly nonlinear systems. Additionally, combining recurrent neural networks with genetic algorithms offers a valuable and cost-effective strategy for enhancing biotechnologica