Fuzzy modelling: Fundamentals, construction and evaluation (Q809959)

Fuzzy sets used in system modelling give rise to a broad class of fuzzy- set based models. Despite different techniques applied to modelling (e.g., probability, fuzzy sets, set theory etc.), the results of it should be evaluated uniformly with respect to some essential properties. Following this line, the main goal of the paper was to present a global methodological overview of fuzzy modelling and fuzzy models. Firstly, essential reasons underlying the need for fuzzy set-based modelling are enumerated and illustrated with the use of several examples. The list of primary requirements is provided. Secondly, the most essential drawbacks of the currently existing methodology is clearly pointed out. The proposed approach introduces a validation mechanism which leads to construction of interval-valued fuzzy sets. By doing that one can easily express the validity of the fuzzy model. The intervals for grades of membership functions characterizing the results of modelling are derived within a process of model building and, in their essence, possess the same flavor as confidence intervals developed in statistical models. The width of the interval is directly implied by the performance of the constructed model and expressed with the use of the following equality index \(a\equiv b=1/2((a\to b)\bigwedge (b\to a)+(\bar a\to \bar b)\bigwedge (\bar b\to \bar a))\) where \(\to\) stands for multivalued implication induced by a certain t-norm, overbar denotes complement (i.e., \(\bar a=1-a)\) and \(\bigwedge\) reads as minimum. Finally, the paper is devoted to a certain class of fuzzy models constructed within a framework of fuzzy relational equations. In this case the detailed identification algorithm is also provided.