查詢結果分析
來源資料
頁籤選單縮合
題名 | 回饋式類神經網路於河川流量推估之應用=Recurrent Neural Network for Streamflow Estimation |
---|---|
作 者 | 張斐章; 黃浩倫; 張麗秋; | 書刊名 | 農業工程學報 |
卷期 | 47:2 2001.06[民90.06] |
頁次 | 頁32-39 |
分類號 | 351.819 |
關鍵詞 | 回饋式類神經網路; 動態神經元; 即時學習演算法; 降雨一逕流歷程; Recurrent neural network; Dynamic neurons; Real-time recurrent learning; Rainfall-runoff processes; |
語文 | 中文(Chinese) |
中文摘要 | 類神經網路可區分為前饋式(Feedforward)及回饋式(Recurrent),前饋式的 各類演算法中又以 Rumelhart 所提出的「倒傳遞類神經網路」( Back-propagation Neural Network,BPN )最具代表性,並有許多成功應用的實例。然而,在預測水文時間序 列時,BPN 通常將時間序列以區塊式輸入(相連時間的輸入串成一輸入向量),以建立輸入 ﹣輸出間的對應關係,預測時則由「樣本比對」方式進行,因此在不同時間點預測相同的輸 入資料時,BPN 會產生相同的輸出,無法有效抓住水文序列的時變特性。 回饋式類神經網路( Recurrent Neural Network, RNN )藉由動態神經元( dynamic neurons ),有效學習時間序列的前後關係,並儲存早期的資訊留到以後使用。即時學習演 算法( Real Time Recurrent Learning )的特性是不需要有大量的歷史資料作為訓練範例 ,能隨真實環境降雨逕流特性的改變作有效而迅速的學習。本研究結合回饋式類神經網路與 即時學習演算法以發展一水文推估模式,並運用於大甲溪上游流量推估,其試驗測試效果良 好,顯示即時回饋式類神經網路的優越能力。 |
英文摘要 | Based on network connection, two types of neural networks architecture are defined: feedforward and recurrent. For feedforward, the Back-propagation Neural Network (BPN) proposed by Rumelhart is probably the most famous algorithm with many satisfied applications. However, the BPN is based on the batch learning process that update action takes place only after the whole training data set has been presented. Apparently, it could not grasp the time variate acharacterists of hydrological time series. This research presents an alternative approach of the Artificial Neural Network (ANN) model to estimate streamflow. The architecture of Recurrent Neural Network (RNN) that we used provides a representation of dynamic internal feedback loops in the system to store information for later use. The Real-Time Recurrent Learning (RTRL) algorithm is implemented to enhance the learning efficiency. The main feature of the RTRL is that it doesn't need a lot of historical examples for training. Combining the RNN and RTRL to model watershed rainfall-runoff processes will complement traditional techniques in the streamflow estimation. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。