Qualitative fuzzy system identification of complex dynamical systems

Fuzzy systems have been proved to be excellent candidates for system dynamics identification. However, they are affected by two drawbacks: the resulting nonlinear model (i) does not guarantee that the generalization property holds unless a large amount of samples is employed, and (ii) is not understandable from a physical viewpoint. These drawbacks are particularly serious when fuzzy identification deals with complex natural systems as the observational data set and/or empirical knowledge can occur to be inadequate. For these systems, the available knowledge of the underlying mechanisms is qualitative and highly incomplete, and does often prevent from formulating a quantitative differential model but not a qualitative one. This paper demonstrates that Qualitative Reasoning methods properly integrated with fuzzy systems yield a hybrid system identification method that overcomes the problems outlined above.