Increasing the predictivity of kinetic models for high-cell-density cultivations

In this work, an optimization-based approach is presented which recognises the switching to new parameters or even to a different model at a certain growth rate improving the quality of model prediction for different time horizon lengths. For the dynamic automatic adjustment to changing kinetics, a moving horizon estimator (MHE) is used. Experimental data from cultivation of Ustilago maydis are used for the model-based parameter identification. The embedded MHE was successfully applied to predict changes in biokinetic constants during membrane bioreactor (MBR) fermentation when very low growth rates and therefore changes in metabolism occur. Setting suited horizon lengths and parameter bounds were found to be crucial for convergence and parameter estimation. The expected drop in maintenance parameters at low growth rates was confirmed when using an optimum number of data points.