Abstract
In case that the information of the structure of the system (i.e. order of the system) is not available, a designer must find a suitable structure to represent the system behavior appropriately. Generally speaking, this problem results in a large scale combinatorial search problem. One of the efficient way to solve this kind of problem is to utilize a so-called evolutionary computation. In this paper, we propose a new modeling method for structureunknown multi-input single-output linear dynamical systems using evolutionary computation. However, even if we use the evolutionary computation, an unreasonably long computational time is often required because of its vast search space. In order to overcome this problem, we try to set the limit of the search space based not on the knowledge of the plant but on AIC (Akaike's Information Criterion). After this process, we apply a genetic algorithm to find a sub-optimal model structure. We also carry out some numerical experiments to show the usefulness of our proposed method from viewpoint of both validity of the structure and accuracy.
