An adaptive method for constructing fuzzy rules

Describes extensions to the recursive partitioning method of constructing rules in a fuzzy system. The method involves recursively breaking a region to be approximated into finer and finer sub-regions, each of the sub-regions then being approximated by a single fuzzy rule. This decomposition process can be determined by means of the required accuracy. We present a way of merging the rules, to adapt them to new situations (data becoming out-of-date, changing requirements for efficiency and accuracy, etc.). We apply the method to approximate a 3D surface and show that it can be an effective approximation method.