The simulation of dynamic processes based on rigorous modeling confronts the problem that, due to the physical foundation of the model, the choice of the mechanistic model to be used for simulation is mostly based on the prevailing physical process phenomena. Typically, these phenomena, which dictate the behavior of the dynamic process, change also over time. Consequently, the mechanistic model used to describe the dynamic process should change respectively. Many methods have been applied for the detection of changing behaviors in dynamic processes, the most representative are the Hybrid state observer in model-based process control and the hidden Markov model  for statistical modeling.
In this work, a novel approach is proposed to achieve an efficient mechanistic simulation of dynamic processes and to gain knowledge about non measurable characteristics of dynamic processes. First, all the mechanistic models, which are proposed for the process considered, are taken into account and analyzed. Next, the process time horizon is divided in a number of intervals defined by the differentiability and the identifiability of the process and the models considered. Afterwards, model discrimination and parameter estimation are carried out in each interval so as to determine, which model suits best. Moreover, points in time at which a model-switch takes place are shifted in order to find its optimal allocation. Finally, all the intervals, which share a border and are described by the same model, will then be merged. The developed approach is applied to enable an accurate simulation of the inhibition of the cellular capacities for glucose uptake and respiration in Escherichia coli fed-batch fermentations and to determine the optimal function of the “product formation rate”. Fed-batch fermentations are characterized by its dynamic behavior. Growth rate, substrate concentration and cellular metabolic activity are a few examples. The growth inhibition effects caused by the strong expression of recombinant proteins are described by applying a sequence of models proposed by Neubauer et all . The candidate models are: overflow metabolism model, a citric acid cycle model, and a maintenance model. By means of the proposed model-sequence, acetate formation and cell density are predicted with a higher accuracy and the decrease of the glucose and respiration uptake capacity is analyzed.
Acknowledgment: The authors acknowledge support from the Cluster of Excellency “Unifying Concepts in Catalysis” coordinated by the Berlin Institute of Technology and funded by the German Research Foundation – Deutsche Forschungsgemeinschaft.
1. Hofbaur, M. and B. Williams, Mode estimation of probabilistic hybrid systems. Hybrid Systems: Computation and Control, 2002: p. 81-91.
2. Leggetter, C.J. and P.C. Woodland, Maximum likelihood linear regression for speaker adaptation of continuous density hidden Markov models. Computer speech and language, 1995. 9(2): p. 171.
3. Neubauer, P., H.Y. Lin, and B. Mathiszik, Metabolic load of recombinant protein production: Inhibition of cellular capacities for glucose uptake and respiration after induction of a heterologous gene in Escherichia coli. Biotechnology and Bioengineering, 2003. 83(1): p. 53-64.
See more of this Group/Topical: Computing and Systems Technology Division