Submitted to IEEE Transactions on Fuzzy Systems
(December 2000), IEEE.
Abstract
Traditionally, the design of fuzzy systems has
relied on expert knowledge. Thus, automation of fuzzy
system synthesis has been long sought after. An
abundance of methods exist for automated fuzzy system design.
Particular attention has been given to soft computing
approaches such as neural networks and evolutionary
computation. However, the majority of these methods either are
partially automated (i.e., rules are constructed given the
fuzzy partitions) or have a fixed number rules. A novel
evolutionary computation approach is developed here to
address the aforementioned drawbacks that also has the added
benefit of finding a locally Pareto optimal set of solutions.
The new approach is tested on two data sets with promising
results.