system identification, multimodels, fuzzy logic, type-2 fuzzy logic, ANFIS, Neuro-fuzzy.

This paper proposes a technique for the identification of non-linear dynamics based on modified ANFIS neuro-fuzzy networks using fuzzy type-2. In this technique, the Takagi- Sugeno(-Kang) structure is modified so that the antecedents of the rules cover concepts of fuzzy type-2, which consist of the inclusion of footprint of uncertainty in membership functions. In pararel, the rule structure is modified to use local models as consequence.

Unlike the current technique of ANFIS, the proposed technique represents the dyna- mics in several regions well defined by local models that, in turn, will have their activation determined through the fuzzy relations of pertinence and rules in which antecedents and consequents are based on fuzzy type-2 logic. The proposed methodology combines the concepts of conventional fuzzy logic with learning strategy using artificial neural networks with the treatment of uncertainties that fuzzy type-2 logic allows. In this con- text, the main techniques of system identification will be presented, as well as the state of the art of artificial intelligence.

A case study was carried out using a thermoelectric system addressed in different works. It is possible to compare the identification techniques using artificial intelligence and the classic NARMAX technique and the least squares method. The intelligent techniques used were two adaptive neuro-fuzzy networks, which use both conventional and logical (fuzzy) type-2 logic. The preliminary results obtained show the advantage and the improvement that can be obtained using an advanced technique compared to the classic method