Windows Programming of Feedforward Neural Network by Quasi-Newton Training Method

• Main Authors: Chin-Pei Lin, 林金貝
• Other Authors: 林顯聖
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/94n5zd

碩士 === 國立臺北科技大學 === 化學工程研究所 === 95 === Recently, Neural network has been applied to many engineering fields. People apply Neural network in psychology、 statistics and aerography. Developing a free and practical windows program of feedforward neural network is the object of our laboratory. The training methods used in the former studies are the momentum backpropagation method and Conjugate Gradient Method. However the training rate of those methods is slow and it must take much time to reach convergence. Quasi-Newton training method is studied to speed up the training rate in this research. Visual Basic 2005 is used to develop the window-based program. Quasi-Newton method is used to train weights and biases of the neural network. In the program, Davidon-Fletcher-Powell method was used to calculate the quasi Hessian matrix. Some stop criterions are added so the programming can stop automatically. Three simulation cases are used to test the applicability of the program. The convergence rate of three training methods was studied . These results show that the program is practicable and the convergence rate of Quasi-Newton methods is the fastest.