整合系統識別、圖形識別與振動控制之類神經網路平台之研究

Development of an Integrated Neural Network Platform for System Identification, Pattern Recognition and Control

陳春志(C. J. Chen) ； 楊世銘(S. M. Yang) ； 黃威龍(W. L. Huang)

《修平學報》 33期 (2016/09) Pp. 79-122

摘要

巨本文發展一個類神經網路整合平台以應用於系統識別、追蹤與減振控制和圖形辨識。此類神經網路整合平台將結合類神經網路理論、基因演算法並建構於個人電腦之視窗介面。研究目的在開發一套可應用於動態系統識別和控制之整合性的類神經網路視窗軟體。此類神經網路軟體（NeuralLink）包括零階、一階和二階等學習法則，如倒傳遞演算法、倒傳遞演算法加慣性項和適應性學習速率法、共軛梯度學習法、馬快特（Marquardt）演算法、擬牛頓學習法和實數型基因演算學習法等，並結合基因演算法以有效加快收斂。此整合平台將以實驗驗證不同網路架構與適配之訓練方法，實驗標的物軟體測試的結果顯示此類神經網路視窗軟體可成功應用於系統識別、字型辨識與定位平台（Shaking Table）之減振控制。

關鍵字

類神經網路 ； 系統識別 ； 圖形識別 ； 減振控制

並列摘要

This paper aims at developing an integrated design and synthesis platform for applications in system identification, vibration and regulation controller design, and pattern recognition. The objective is to provide a platform based on neural networks integrating with neural networks and genetic algorithms such that design and synthesis of neural network can be conducted effectively in pc/windows environment. For system dynamics varying from one case to the other, the training algorithms include zero-order, one-order and two-order are necessary. In addition to the training algorithms currently available, the backpropagation algorithm, the backpropagation learning algorithm with momentum term and adaptive learning rate, Gauss-Newton algorithm, conjugate gradient algorithm, Levenberg-Marquardt algorithm Quasi–Newton (DFP) algorithm, Quasi–Newton (BFGS) algorithm, and the integrated floating genetic algorithm are developed for accelerated training efficiency. Examples are employed to demonstrate the effectiveness of training algorithms, and experiments on system identification, vibration suppression and tracking control will also be conducted. All of the above function can be integrated in the platform so that no programming is required in engineering applications.

並列關鍵字

Neural Network ； System Identification ； Pattern Recognition ； Vibration Suppression

參考文獻

Jacobs, R. A. (1988). Increased Rates of Convergence Through Learning Rate Adaptation. Neural Networks, 1: 295-307.

Baba, N. (1989). A New Approach for Finding the Global Minimum of Error Function of Neural Networks. Neural Networks, 2: 367-373.

Nguyen, D., and Widrow, B. (1990). Improving the Learning Speed of 2-Layer Neural Networks by Choosing Initial Values of the Adaptive Weights. Proc. of IJCNN,San Diego, California, 3: 22-26.

Minai, A. A., and Williams, R. D. (1990). Acceleration of Back-propagation through Learning Rate and Momentum Adaptation. Proc. of IJCNN, 1: 676-679.

Yu, X. H., Chen, G. A., and Cheng, S. X., (1995). Dynamic Learning Rate Optimization of the Backpropagation Algorithm. IEEE Trans. Neural Networks, 6(3): 669-677.