Name
Abstract | ||
|---|---|---|
Process modelling is an important discipline both in science and engineering. The complexity of many real world systems has lead to sophisticated modelling approaches, where both the accuracy and readability of the models are of great importance. Fuzzy modelling is such an approach, which uses well-established machine learning techniques, producing models with the capacity of integrating expert knowledge with real world observations. The behaviour of these models is described as a series of linguistic rules. The readability of the models is related to the number of rules used to describe the system. In order to define methodologies for organising the information describing a system, it is important to define metrics for the relative importance of a set of rules in the description of a given region of the input/output space. This paper addresses this problem, and a new concept is proposed: the relevance of a set of rules. The concept is defined by a set of intuitive axioms, leading to a set of properties that any function of relevance must obey. In order to corroborate the validity of the new concept, a new methodology is presented. It has been called SLIM (Separation of Linguistic Information Methodology). It is useful for organising the information in a fuzzy system: a system f(x) is organised as a set of n fuzzy systems f(1)(x), f(2)(x), ..., f(n)(x). Each of these systems may contain information related with particular aspects of the system f(x). The application of the SLIM methodology in a hierarchical structure is described. The structure, HPS (Hierarchical Prioritised Structure), has been first proposed by Yager [1,2,3]. SLIM is used to organise the information in the various layers of the structure. It is also possible to reduce the number of rules representing the original system by discarding rules with lower relevance values. In order to corroborate the proposed concepts, experimental results are presented for the fuzzy identification of a surface with the form of a volcano. The description of this surface by a fuzzy rules system, using a nearest neighbourhood identification method, produces a system containing a high number of rules (400 rules). With the new approach, the surface can be described with only 51 fuzzy rules, in a hierarchical structure. The new system, with three rules, is obtained using SLIM from the initial system with 400 fuzzy rules, without significant loss of information content. To further corroborate the usefulness of the concept of relevance, experiments have been conduced in the identification of environmental variables in a agricultural greenhouse (temperature and humidity). The results of using SLIM in the "organisation" of the identification system are briefly described. A summary of the critical analysis of these results is also presented. They indicate, once more, the validity of the concept. |
Year | DOI | Venue |
|---|---|---|
2000 | 10.1109/ICSMC.2000.886597 | IEEE International Conference on Systems Man and Cybernetics Conference Proceedings |
Keywords | DocType | ISSN |
temperature,input output,artificial intelligence,process modelling,learning artificial intelligence,volcanoes,machine learning,humidity,arithmetic,fuzzy logic,fuzzy systems | Conference | 1062-922X |
Citations | PageRank | References |
4 | 0.90 | 6 |
Authors | ||
4 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Paulo Salgado | 1 | 11 | 2.54 |
| Pedro Melo-Pinto | 2 | 152 | 16.21 |
| José Bulas-Cruz | 3 | 81 | 10.99 |
| c oouto | 4 | 4 | 0.90 |