Neural network modeling and optimization of biotechnology processes.
详细信息
- 作者：Tholudur ; Arun Narayanaswamy.
- 学历：Doctor
- 年：1998
- 导师：Ramirez, W. Fred
- 毕业院校：University of Colorado
- 专业：Engineering, Chemical.;Biology, Microbiology.;Engineering, System Science.;Computer Science.
- ISBN：0599157143
- CBH：9916844
- Country：USA
- 语种：English
- FileSize：5345422
- Pages：191
文摘
Many biotechnology reactor systems involve the growth of cells (bacterial, fungal or mammalian) in either batch, fed-batch or continuous reactors with the aim of obtaining a product made by the organism under consideration. This product could be either important proteins or enzymes. The successful optimization of the process to obtain maximum amounts of the desired product requires a good model of the process. Furthermore, a microscopic model of the process which looks at the cause and effect of every variable would not only be time consuming to develop, but also involve many unmeasurable variables so that the model obtained would not be amenable to optimal control methods. Macroscopic models such as overall kinetics based models require an in- depth knowledge of the process dynamics.;The aim of this work is to provide a framework for using neural networks to model complicated dynamic biotechnology processes. Neural networks have very well known function approximation capabilities. The method of parameter function neural network modeling has been developed in which a priori information about process dynamics is used in conjunction with neural networks to yield a macroscopic model of the process. This process model can then be optimized to obtain optimal operating conditions or optimal control policies.;The method of neural network parameter function modeling is applied to two experimental systems---one involving the production of the protein beta-galactosidase using a recombinant Escherichia coli system, and the other involving the production of cellulase proteins using the fungus Trichoderma reesei.;For the recombinant E. coli system, shake flask experiments are carried out to generate data. Neural network parameter function models and kinetic models for growth and protein production dynamics are developed. Optimal conditions for operation of the process are identified and verified.;Fermentations of T. reesei are carried out in 5L bioreactors. Kinetic and neural network parameter function models are developed to describe the growth and cellulase production characteristics of the fungal system. Based on the models developed. optimal conditions are obtained.;The neural network parameter function modeling technique has been experimentally shown to be a powerful technique for modeling dynamics of non-linear processes.