Digital Circuit Design for a Fuzzy Neural Network and Dynamic Parameter Learning Rule

• Main Authors: Wen-Pin Chu, 朱文彬
• Other Authors: Chiang-Ju Chien
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/74587027317225028438

碩士 === 華梵大學 === 電子工程學系碩士班 === 100 === The parameter learning algorithm for a fuzzy neural network is a long term study issue in the area of intelligent control. In this thesis, we investigate the backpropagation dynamic parameter learning algorithm for the fuzzy neural network with triangular fuzzy sets. We derive the parameter learning rules for the mid-point and width of the triangular fuzzy set as well as the output weighting parameters. A set of input output data of a third order nonlinear system is utilized as the training data for computer simulation. A MAtlab program is written to demonstrate the learning performance and prove the correctness and effectiveness of the learning algorithm. The main contribution of this thesis is to implement the fuzzy neural network and backpropagation dynamic parameter learning algorithm as a digital circuit. The circuit can be realized by using the basic adder, subtractor, multiplier and divider circuits since the triangular fuzzy set is adopted. The whole digital circuit can be divided into four parts including the fuzzy neural network circuit, backpropagation dynamic parameter learning circuit, training dtat ROM circuit and control unit circuit. In addition to the discussion of the circuit design structure and theory, computer simulation is also made for each important subcircuit and the whole circuit. We use the VHDL to implement the circuit. The circuit is then synthesized by using Altera Quartus II and downloaded into an Altera Cyclone II development kit. The FPGA experimental results and the Quartus II computer simulation results are consistent. This confirms that the circuit design is feasible and correct.