Evolving fuzzy rule-based models

Angelov, Plamen (1999) Evolving fuzzy rule-based models. In: 8th IFSA World Congress, 1999-08-17 - 1999-08-20.

Microsoft Word (ifsa1)
ifsa1.doc - Submitted Version
Download (65kB)

Abstract

An approach to auto-generation of fuzzy rule-based models is proposed in the paper. Its main advantage is the high flexibility: no a priory information about the model structure is necessary. A new efficient coding procedure is proposed instead of usually used coding of all possible fuzzy rules into a chromosome. Parameter and structure identifications are realized by optimization at two stages. First, the best k rules are determined such that to minimize the root square error. At the second stage, GA tunes parameters of the membership functions and singletons. Software, which realizes the approach in the framework of Matlab v.5.2, is designed. Modeling of thermal load of a building is considered in order to illustrate the applicability of the approach.