Classification and clustering of fuzzy rule

Information Methodology (SLIM), has been proposed [Salgado 1999]. This methodology is inspired in the HPS methodology firstly proposed by Yager [1993a, 1993b, 1998].
Preliminary experiments using SLIM produced very good results.
This methodology has been applied to the identification of the environmental variables in a greenhouse situated in the UTAD campus [Salgado 1997, 1998, 1999] and to the identification of the model of a chemical process carried out in a pilot plant existent in the Chemical Engineering Department of Coimbra University [Salgado et. al., 2001].
Fuzzy identification systems for this purpose, created with established unsupervised methods, are made of a very high number of IF…THEN rules. By applying SLIM it was possible to organize the fuzzy rules in different hierarchical structures, obtaining then a reduction between 80% and 98% on the number of this rules, with negligible differences in the behavior of the systems.
The definition of distance among rules creates a metric space of the rules. It opens the doors to the development of new methodologies to analyze fuzzy systems. Therefore, the applicability of the SLIM has been increasing as well as the classification of the fuzzy rules, namely in terms of its clustering [Salgado 2001]. This new metric leads to the development of different fuzzy rules operations, and therefore to its handling and interpretation.
In this project, the above methodology will be refined and tested, in order to further corroborate it. In the scope of the project two application areas will be explored: fuzzy identification systems and fuzzy patterns recognition. These two areas have well-established methods, which are appropriate to test and critically evaluate the concept of relevance with the associated SLIM and the fuzzy distances.