Self-Learning Fuzzy Logic Controllers Design

• Main Authors: Ming-Po Chuang, 莊明博
• Other Authors: Chung-Chun Kung
• Format: Others
• Language: en_US
• Published: 1993
• Online Access: http://ndltd.ncl.edu.tw/handle/57868009638572216335

碩士 === 大同工學院 === 電機工程研究所 === 81 === The main purpose of this thesis is to utilize the error-backpropagation algorithm of the conventional neural network to design a fuzzy logic controller (FLC), which is a form of rule-based controller. The main advantage of the proposed FLC design is that the design procedure is simple and effective, and we can avoid some complexity and difficulty while designing nonlinear control systems using conventional methods. Another advantage of this design method is that we can choose some representative training patterns to design fuzzy logic controllers and accquire robustness to some extent by generalization property. The control rule used in FLC is represented as ─ Ｒ: IF (premise clause), then (consequence clause). The premise of a rule is the description of fuzzy subspace of inputs and its consequence is a input-output relation. In this thesis, we propose two modified forms of consequence clause, which are nonlinear input-output relations and different from the conventional linear form. One is in the form of two variables' polynomial of first order and the other is in the form of nonlinear multilayered neural network. For illustration, we design fuzzy logic controllers for two nonlinear systems. One is the inverted pendulum system of second order, and the other is a typical nonlinear system with unity gain. For the purpose of practicality, we establish a decision table from the well-trained fuzzy logic controller about the inverted pendulum system's model to control the physical inverted pendulum system.