Recurrent Fuzzy Neural Network Controller Designed by Dynamic Backpropagation Algorithms and Implemented in MRAS

• Main Authors: Jung-Kuei Tsai, 蔡榮桂
• Other Authors: Lin-Ying Lai
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/23231625762159366468

碩士 === 中原大學 === 電機工程學系 === 88 === In the thesis, a recurrent fuzzy neural network controller (RFNNC) is proposed and realized in a model reference adaptive system (MRAS). The proposed controller combines the recurrent neural network with fuzzy logic control. The memory element is represented by adding feedback connection in feedforward neural network. The parameters of the fuzzy model are tuned on-line by a generalized dynamic backpropogation algorithm. Compared to a common fuzzy neural network controller, the proposed RFNNC owns two characteristics. Firstly, the structure is inherently a recurrent multilayered network for realizing dynamic fuzzy logic inference. The recurrent property makes it suitable for dynamic problems. Secondly, not only the identification process can be omitted, but also the plant output information along is required. Besides, by the selected cost function, the simplified sensitivity function, fewer learning parameters are needed for realization. The proposed RFNNC is applied to the simulation of a second order linear system, a nonlinear system, a highly nonlinear system and a non-BIBO (non bounded-input bounded-output) system. The simulation results show that the controlled systems have good tracking ability even for instant-varying trajectories. And the adaptation ability for instant load is also good.