Rule-based neuro-fuzzy modelling of dynamic system and designing of controllers

Abstract

Models of dynamic systems are necessary, for instance, in simulation, prediction, model-based control and fault diagnosis. System modelling based on conventional mathematical tools (e.g., linear or nonlinear differential or difference equations), yielding quantitative numerical models, is not well suited for dealing with ill-defined, complex and uncertain systems. On the other hand, fuzzy modelling employing fuzzy IF-THEN rules, provides a tool for designing qualitative models without employing precise quantitative analyses. However, there are many situations where expert domain knowledge, which is usually the basis for designing fuzzy models, is not sufficient, due to incompleteness of the existing knowledge, problems caused by different biases of human experts, difficulties in forming rules, etc. For this reason, methods for data-driven fuzzy modelling and identification are of great interest. Among them, methods from the field of computational intelligence (CI) take a remarkable place. This is mainly because they are effective tools for designing “intelligent” models, that is, models that are able to learn from examples (described by both numerical and linguistic fuzzy data), to generalize from the learned knowledge and to explain the actions they make.