On-Line Application of Parameter Estimation Accuracy to Biotechnical Processes

Many biotechnical processes can be modelled in principle, but the model parameters are not known exactly because of changing media composition or strain variability. Therefore, the application of filter, control or prediction algorithms demands for a repeated parameter estimation from time to time during a fermentation run. But not all of the parameters of a complex model can be estimated on-line at each particular fermentation time on the basis of the data sampled so far, because the process may show different physiological states like substrate saturation or substrate limitation or measurement accuracy and sampling rates are simply to small. The evaluation of the eigenvectors and eigenvalues of the Fisher information Matrix allows for the calculation of the parameter estimation accuracy. This leads to an optimal selection of the parameters suitable to be identified on-line. An improved predicton of the process can be obtained by choosing the nominal values for the parameters that can be identified and the estimated values for the parameters that have already significant influence on the process.