Abstract:
Reinforcement learning refers to a class of learning tasks and algorithms in which the learning system learns an associative mapping by maximizing a scalar evaluation fun...Show MoreMetadata
Abstract:
Reinforcement learning refers to a class of learning tasks and algorithms in which the learning system learns an associative mapping by maximizing a scalar evaluation function by interacting with environment. Fuzzy actor critic learning (FACL) is a reinforcement learning method based on dynamic programming principle. A priori knowledge of the process either in the form of models or experts is required for tuning the conclusion part of the fuzzy inference system (FIS). This paper proposes a novel algorithm using FACL to automatically tune the conclusion part of the FIS. The only information available for learning is the system feedback that describes the rate of reward or punishment for the action performed at the previous state. Reinforcement learning problems are discrete time dynamic problems, in which the learner has classically a discrete state perception and trigger only discrete actions. It is planned to apply the same for the control of continuous processes. The generality of these methods allows the system to learn every kind of reinforcement learning problems. The experimental studies of these methods have also shown superiority of these methods over the related reinforcement methods as stated in the literature. In this paper, the proposed reinforcement learning algorithm was initially applied to boiler drum level system and the performance was studied. To show the generality of these methods, it was also applied to a number of linear processes.
Published in: 2007 American Control Conference
Date of Conference: 09-13 July 2007
Date Added to IEEE Xplore: 30 July 2007
ISBN Information: