Neuro-fuzzy modelling-architecture and modelling issues

摘要

The paper aims at discussing neuro-fuzzy models; modelling and identification issuers useful in practical applications in process control, decision making and classification, when modelling system static and dynamic characteristics. Novel modification of Mamdani's inference algorithm is described to increase the model interpretability. General neuro-fuzzy model in the form of connectionist architecture suitable for parallel computing is revealed. Role os constraints in maintaining the interpretability of a fuzzy system and its model is discussed.