|||Gerhard Goos, Martin Spott, Joachim Weisbrod, Martin Riedmiller, Interpretation und Adaption unscharfer Relationen innerhalb einer hybriden selbstlernenden Steuerungsarchitektur, Proc. in AI: Fuzzy-Neuro-Systems '97, p. 332-339, Infix-Verlag, Mar 1997.
In this paper the hybrid, selflearning control architecture Fynesse is introduced, which consists of an adaptive fuzzy controller and a neural critic element. The neural critic is trained by methods of dynamic programming and evaluates the control strategy of the fuzzy controller, which will be adapted according to this information. The advantages of neural techniques (learning capability and distributed information representation) and fuzzy techniques (interpretability and possibility to include uncertain a priori knowledge) are preserved, since these components are strictly separated. This fact - in addition to the capability of being selflearning - is one of the main properties that distinguishes Fynesse from other Neuro-Fuzzy-Systems. In this paper we focus on the fuzzy controller and especially its interpretation.