(574p) Data Analysis and Parameter Estimation of Bioprocesses

The estimation of state variables and model parameters based on a rigorous process model together with measurement data is a key step in model-based process applications. Models which describe fermentation processes show high non- linearity and time variant kinetic parameters. However, a good quality of parameter estimation enables the realization of well posed and effective control strategies. Thus, the potential to evaluate the quality of the parameters can be increased based on practical identifiability analysis i.e. using real noise-corrupted experimental data series. In this work, the dynamical behaviour of the parameter system is taken into account. One requirement for a successful state or parameter estimation is that the system is in general identifiable or observable. Therefore, in this work an approach is applied using the analysis of the quantitative identifiability, since measurement errors are considered. Moreover, in order to determine the quantitative identifiability, the confidence region of the parameter estimates has been analyzed. Here, an asymptotic lower bound of this matrix can be given by the inverse of the Fischer Information matrix (FIM). To scale the different values in the objective function and to consider the accuracy of the measurement devices, the inverse of the variance-covariance matrix of the measurement errors is used as a weighting in the objective function. The proposed optimization-based approach has been applied to improve the predictivity of kinetic models based on available measurements together with a process model. In this study, experimental cultivation data are used for model-based parameter identification.