Construction of fuzzy rule base using hinging hyperplanes algorithm from training data

This paper describes a new method for automatically learning the rules of a fuzzy system from numerical training data. The method allows one to approximate the training data by a fuzzy model. It is suitable for fuzzy modeling and model-based control. The method consists of two steps. In the first step, the data are approximated by a set of hyperplanes using the ´hinging hyperplanes´ algorithm. In the second step, the hyperplanes are described by fuzzy IF-THEN rules. The algorithm allows one to incorporate expert knowledge about the system.