Keywords
genetic algorithms, real-coded, robust experimental design
Start Date
1-7-2006 12:00 AM
Abstract
When calibrating a (dynamic) model, one is often faced with a lack of information-rich data. Without such data, there is little hope in obtaining accurate parameter estimates. In order to improve the situation, optimal experimental design for parameter estimation (OED-PE) can be employed. The main drawback of the classical OED-PE methodology is that values for the model parameters need to be provided in order to obtain an optimal design. If the values of the model parameters are highly uncertain, robust OED-PE should be preferred, yielding a design which guarantees a certain information content given the parameter uncertainty. This approach adds another level of optimization to the design problem. For each proposed experiment (optimization of the experimental degrees of freedom) an additional optimization covering the whole parameter domain needs to be performed. In this work the maximin robust OED-PE technique will be illustrated with a simple model describing substrate consumption based on Monod kinetics. The optimization problem consists of two nested real-value genetic algorithms in which each fitness evaluation for the optimization of the experimental degrees of freedom requires a full genetic algorithm optimization over the parameter domain.
Nesting genetic algorithms to solve a robust optimal experimental design problem
When calibrating a (dynamic) model, one is often faced with a lack of information-rich data. Without such data, there is little hope in obtaining accurate parameter estimates. In order to improve the situation, optimal experimental design for parameter estimation (OED-PE) can be employed. The main drawback of the classical OED-PE methodology is that values for the model parameters need to be provided in order to obtain an optimal design. If the values of the model parameters are highly uncertain, robust OED-PE should be preferred, yielding a design which guarantees a certain information content given the parameter uncertainty. This approach adds another level of optimization to the design problem. For each proposed experiment (optimization of the experimental degrees of freedom) an additional optimization covering the whole parameter domain needs to be performed. In this work the maximin robust OED-PE technique will be illustrated with a simple model describing substrate consumption based on Monod kinetics. The optimization problem consists of two nested real-value genetic algorithms in which each fitness evaluation for the optimization of the experimental degrees of freedom requires a full genetic algorithm optimization over the parameter domain.
