Rule extraction using a novel class of fuzzy degraded hyperellipsoidal composite neural networks
Mu‐Chun Su, Chien-Jen Kao, Kai-Ming Liu, Chi-Yeh Liu · 2002
Presents an innovative approach to rule extraction directly from experimental numerical data for system identification. The authors discuss how to use a novel class of fuzzy degraded hyperellipsoidal composite neural networks (FDHECNN's) to extract fuzzy if-then rules. The fuzzy rules are defined by hyperellipsoids of which principal axes are parallel to the coordinates of the input space. These rules are extracted from the parameters of the trained FDHECNN's. Based on a special learning scheme, the FDHECNN's can evolve automatically to acquire a set of fuzzy rules for approximating the input/output functions considered systems. A highly nonlinear system is used to test the proposed neuro-fuzzy systems.>