Fuzzy controller generating procedures when using crisp inputoutput data produce the necessary system in two steps: first they produce a starting rule set and then they tune the parameters that influence the approximation with a learning algorithm. Other solutions work under special conditions as hybrid neuro-fuzzy systems improving the approximation with a gradient based learning algorithm (e.g. in the case of monotonous membership functions), or use the methods of the genetic algorithms to generate the fuzzy controller. This article demonstrates a new method which reduces the problem to a classification task and carries out the generation of the rules and the tuning of the system in a single step.
展开▼
