A neuro-fuzzy approach to optimize hierarchical recurrent fuzzy systems

To simplify the definition of fuzzy systems or to reduce its complexity hierarchical structures can be used. Thus, more transparent rule bases that are also easier to maintain can be designed. Furthermore, it is sometimes necessary to use time delayed input or to reuse time delayed output from the fuzzy system itself to obtain a rule base that describes the analyzed problem appropriately. This leads to hierarchical recurrent architectures that have increased approximation capabilities since they are able to store information of the past. In this article we present a neuro-fuzzy model that can be used to optimize hierarchical recurrent fuzzy rule bases if training data is available. Furthermore, we present an approach to learn initial rule bases from data using rule templates. © 2002 Kluwer Academic Publishers.