Fynesse: a hybrid architecture for self-learning control

|A system is presented that autonomously learns to control a priori unknown nonlinear dynamical systems and enables the user to interpret, examine and correct the control strategy in every stage of learning. Based on ve requirements that are essential to us for a widely applicable learning controller the hybrid control architecture Fynesse is derived: The control strategy is represented by a fuzzy relation that can be interpreted and contain a priori knowledge, whereas the more complex part of learning is solved by a neural network that is trained by dynamic programming methods. The advantages of both paradigms | learning capability of neural networks and interpretability of fuzzy systems | are preserved since the two modules are strictly separated. The application of the Fynesse controller to a chemical plant demonstrates the high quality of autonomously learned control. Furthermore it shows the beneets of the integration of a priori knowledge and the interpretation of the controller in terms of fuzzy rules.

[1]  C. Watkins.  Learning from delayed rewards , 1989 .

[2]  Paul J. Werbos,et al.  Overview of designs and capabilities , 1990 .

[3]  Hamid R. Berenji,et al.  Learning and tuning fuzzy logic controllers through reinforcements , 1992, IEEE Trans. Neural Networks.

[4]  Nadine N. Tschichold-Gürman,et al.  FUN: optimization of fuzzy rule based systems using neural networks , 1993, IEEE International Conference on Neural Networks.

[5]  Andrea Bonarini,et al.  A simple direct adaptive fuzzy controller derived from its neutral equivalent , 1993, [Proceedings 1993] Second IEEE International Conference on Fuzzy Systems.

[6]  Frank Klawonn,et al.  Combining Neural Networks and Fuzzy Controllers , 1993, FLAI.

[7]  J. Sitte,et al.  A cartpole experiment benchmark for trainable controllers , 1993, IEEE Control Systems.

[8]  Li-Xin Wang,et al.  Adaptive fuzzy systems and control , 1994 .

[9]  Joachim,et al.  Fuzzy Control Revisited | Why Is It Working ? , 1995 .

[10]  Andrew G. Barto,et al.  Learning to Act Using Real-Time Dynamic Programming , 1995, Artif. Intell..

[11]  Martin A. Riedmiller.  Learning to Control Dynamic Systems , 1996 .

[12]  John N. Tsitsiklis,et al.  Neuro-Dynamic Programming , 1996, Encyclopedia of Machine Learning.

[13]  Martin A. Riedmiller,et al.  Application of sequential reinforcement learning to control dynamic systems , 1996, Proceedings of International Conference on Neural Networks (ICNN'96).

[14]  Spott Joachim WeisbrodInstitut.  A new approach to the adaptation of fuzzy relationsMartin , 1996 .