The study of automatic insertion and deletion of fuzzy rules in fuzzy neural network models

This research is based on a fuzzy neural network, named knowledge-based neural network with trapezoid fuzzy set inputs (KBNN/TFS). We use this network model to refine fuzzy rules with a training database. We propose an interactive consistency checking engine with automatic rule insertion and deletion (ICE/RID) to perform fuzzy rule verification. This process is used to verify the initial rule base and the rules refined by KBNN/TFS. With the interactive interface of ICE, we can detect redundant rules, subsumed rules, and conflict rules. Besides, we can also use RID to insert and delete fuzzy rules automatically if necessary. The proposed model is tested with an inverted pendulum system (IPS). In these experiments, we demonstrate the ability of ICE/RID to remove inconsistencies and insert rules in KBNN/TFS. With the combination of ICE/RID and KBNN/TFS, a valid and consistent rule base can be obtained.