MATLAB/Simulink as a Framework for Meta-Model based Evaluation of System variants

Olaf Hagendorf¹, Thorsten Pawletta¹, Roland Larek²

¹Hochschule Wismar, Phillip-Mueller-Strasse / PF1210, 23952 Wismar
²Stiftung Institut für Werkstofftechnik IWT, Badgasteiner Str. 3, 28359 Bremen

Modeling and simulation with integrated parameter optimization is used routinely to
improve system performance. In engineering a well-known environment for this task is
MATLAB/Simulink. Using this or similar, established techniques only model parameter
values are optimized. Model structure is considered to be fixed. Until now no methods are
known which can optimize systems utilizing all degrees of freedom. As system
performance is optimized it may be necessary to redesign the model structure. This is
carried out manually by an analyst but not automatically by the optimization. The
suboptimal combination of automatic parameter optimization and manual structure
changes leads to an error-prone optimization task.
The System Entity Structure/Model Base framework (SES/MB) is able to define
alternative model structures and parameter sets in a single meta-model. Atomic models
are stored in a MB. Using both, it is possible to generate modular, hierarchical models
with different structures and parameters.
Evolutionary Algorithms are a subtopic of Artificial Intelligence that is involved in
combinatorial optimization problems. These algorithms are based on ideas inspired by
biological evolution. They often perform well for many problem types because they do
not make assumptions about the problem specific search space.
The research reported in this paper details an approach providing optimization through
automatic reconfiguration of both: model structure and model parameters. An
evolutionary algorithm based optimization method is assisted by an SES/MB based
model management. It searches for an optimal solution with repeated, combined model
parameter and model structure changes resulting in a combined parameter and structure
optimized model.