Multimodel modelling of compex systems by natural classification

Abstract EN

In this work, we are interested in the modeling of complex systems by means of adopting the multimodal approach.

The use of a natural classification method based on probabilistic neural network "PNN" and the "Kohonen" competitive apprenticeship networks allows determining the representative model library of the system to be modeled.

This classification method requires priority knowledge about the representation space of the collected measures, as well as about the number of classes to generate the number of the base models.

In this paper, we propose to adopt the principal component analysis "PCA" for the selection of the systems measures' projection axes, as well as the algorithm of dynamic clouds "kmeans” for the automatic choice of the number of classes.

The use of these two methods proposes an interesting solution for these two steps which must necessarily precede the classification step.

A comparative study between the two results, by these two neural networks types, is provided in this study while two benchmarks are studied to prove the performance of probabilistic networks for the modeling of complex systems.