Learning Behavior of Fuzzy Controllers with Neuron Adaptive Elements
Yung-Yaw Chen, Kao-Zong Lin · 1992
In this paper, two neuron adaptive elements ae devised to work with a fuzzy control system so that the fuzzy rule-base can be derived through repetitive trials. The discussed architecture is capable of learning and deriving appropriate control actions in a fuzzy control system. To simplify the problem, only the action part of the rule-base is unknown before the learning with the premise part pre-determined. Even with such two simple neuron elements, the scheme successfully presents a trained fuzzy controller which can function independently, i.e. with the learning mechanism detached, after the training. Moreover, a number of different dynamic systems Were tested and the result showed that the proposed algorithm is able to handle more than just the usually seen inverted pendulum.