| Abstract [eng] |
In the Master’s thesis an overview of biotechnological processes, bioreactors and their types is provided, as well as measured variables and the ongoing reactions are described. Further, the equations of mass balance of the processes and reaction rate models are analyzed. Typical structures of hybrid models and their identification methods are provided. The fed-batch process of recombinant protein production was chosen as an object of the investigation. For the research, a mechanistic model of the process and the experimental data from the literature were used. Using the reaction rate equations, mass balance equations and model parameters, the mechanistic model was modified into a hybrid model with an artificial neural network (ANN). The main aim of the study is to identify hybrid model of the process. Identification was carried by using Levenberg - Marquardt algorithm, the results of which were adjusted by sensitivity equation approach. Within the study, it was ensured that the appropriate structure of artificial neural networks is chosen. Training and testing of the artificial neural networks were conducted. The major emphasis was given for ANN model of specific protein accumulation rate, which has two input variables: the specific biomass growth rate and specific protein activity. The conclusion of the study gives a comparison the mean absolute and mean percentage absolute errors of mechanistic and hybrid models. The practical use of the identified model was considered. |
