A system is presented that 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. The application of dynamic programming methods to train a neural network for such control purposes was already quite successful. In order to shorten the long training process the system should allow the incorporation of a priori knowledge about the control strategy in terms of classical linear controllers, fuzzy control rules etc. As a solution, we propose the hybrid control architecture Fynesse: 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 as before. This article reports results on the application of the Fynesse controller to a chemical plant.
[1]
Paul J. Werbos,et al.
Overview of designs and capabilities
,
1990
.
[2]
오경환,et al.
[特輯]퍼지 추론(Fuzzy Reasoning)
,
1992
.
[3]
Ben J. A. Kröse,et al.
Learning from delayed rewards
,
1995,
Robotics Auton. Syst..
[4]
Ivan Bratko,et al.
Learning to control dynamic systems
,
1995
.
[5]
Joachim,et al.
Fuzzy Control Revisited | Why Is It Working ?
,
1995
.
[6]
Andrew G. Barto,et al.
Learning to Act Using Real-Time Dynamic Programming
,
1995,
Artif. Intell..
[7]
Martin A. Riedmiller.
Learning to Control Dynamic Systems
,
1996
.