Design of the Fuzzy Control Systems with Online Learning Ability

• Main Authors: Ling, Yu-Wei, 凌育偉
• Other Authors: Wong, Ching-Chang
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/30366215135656416969

碩士 === 淡江大學 === 電機工程學系 === 86 === 　　In this thesis, fuzzy control systems with online learning ability are proposed. In the fuzzy control system design, reinforcement learning is one of the famous approaches with the online learning ability, but its inherent structure is the trial-and -error type. The trial-and -error type method may be useless for some practical applications owing to its spending too much time. Therefore, we propose a new fuzzy control structure, which has the online learning ability but withdraws the type of trial-and-error. The proposed control system structure can handle the controlled system without knowing its mathematical model under the situations that only the information of the control object and the boundary condition are given. In this structure, the Adaptive Heuristic Critic (AHC) is considered and a state evaluator plays the role of a critic element to point out the performance value of the current state. A functional evaluator and a fuzzy evaluator are individually proposde to produce a scalar value, which is viewed as an internal reinforcement signal for the parameter adjustment. The Temporal Difference (TD) learning is used to adaptively tune the parmeters of the fuzzy controller. The goal of the parameter adjustment is to maximize the evaluation value of the current state such that the control object can be attained. Finally, some results are used to show the robustness and the adaptability of the proposed method.