• 作者： 曾裕強; 陳錕山
• 作者服務機構： 聯合工商專科電子科; 中央大學太空及遙測中心
• 中文摘要： 本文提出一建基於以多項式基底函數模式化之神經網路的動態學習演算法。此一學習演算法採用卡曼濾波器的技術來調整網路的權植，使得輸入資料的統計特性能隱含於網路中。用以表示網路權植學習速率的卡曼增益是以U-D因子法加以計算和修正。本文採用元件參數萃取的問題來做為神經網路應用的數值例證。與倒傳式學習演算法相比，動態學習神經網路不僅提高了學習階段的收斂速度，同時亦改進了具有高度非線性邊界的問題之可區分性。因此，動態學習神經網路在實際應用上確實是一有效的工具。
• 英文摘要： Based on the polynomial basis function (PBF) modeled neural network, a dynamic learning (DL)algorithm is proposed in this paper. The presented learning algorithm makes use of the Kalman filtertechnique to update the network weights, in the sense that the stochastic characteristics of incoming datasets are implicitly incorporated into the network. The Kalman gains, which represent the learning ratesof the network weight updating, is calculated by using U-D factorization techniques. Numerical illustrationsare carried out using a component parameter retrieval problem. As compared to the BP algorithm, theDL neural network not only substantially increases the convergence rate in the learning stage, but alsoenhances separability for highly nonlinear boundary problems. Results indicate that the proposed DL neuralnetwork provides an efficient tool for practical applications.
• 中文關鍵字： neural network; polynomial basis function; Kalman filter technique
• 英文關鍵字： --