查詢結果分析
來源資料
相關文獻
- 類神經網路應用於雙相氣泡流相傳遞特性之即時決定
- Neural Network Procedures for Taguchi's Dynamic Problems
- 應用類神經網路預測東南亞金融風暴期間臺幣對美元的短期匯率
- 製程間具相關數據之趨勢模型偵測探討
- 手寫數字辨識模式之建立--結合遺傳演算法與類神經網路
- Form Segmentation and Component Classification for Clinic Document Image Analysis
- 類神經網路應用於臺灣股市預測:統合基本面與技術面資訊
- 倒傳遞類神經網路在波浪時序列預報之應用
- 類神經網路在集水區降雨逕流模擬之應用
- 類神經網路應用於影像處理魷魚分級之研究
頁籤選單縮合
題名 | 類神經網路應用於雙相氣泡流相傳遞特性之即時決定=Real-Time Determination of Phase Transport Characteristics in Bubbly Pipe Flows with Artificial Neural Networks |
---|---|
作者姓名(中文) | 楊安石; 林柏宏; |
作者姓名(外文) | Yang, An-shik; Lin, Po-hung; |
書刊名 | 大葉學報 |
卷期 | 11:1 2002.06[民91.06] |
頁次 | 頁51-57 |
分類號 | 332.6 |
語文 | chi |
關鍵詞 | 類神經網路; 倒傳遞; 雙相氣泡流; 相分佈; Neural networks; Back propagation; Two-phase bubbly flow; Phase distribution; |
中文摘要 | 雙相氣泡流的物理傳輸機制可經由廣泛地實驗量測與複雜的理論模擬等探討過程而了解;然而,在實際的工業應用中常需要快速地決定紊流氣泡流場動態特性,故而本研究運用類神經網路以即時掌握雙相氣泡流流域內相發展的機制。分析係基於Liu’s所量測實驗的數據資料分別建立三個倒傳遞類神經網路,以預測雙相氣泡紊流的空泡分率、液相速度和氣相速度等分佈。三個訓練完成後的倒傳遞類神經網路計算輸出與原量測數值比較後,驗證結果發現預測和目標向量之均方根誤差低於4.33%。本研究也完整調查了各種網路參數(包含了隱藏層數目、訓練對數目、轉換函數型式、學習增加率數目、學習減少率數目和動量項數目等)對於類神經網路訓練品質的影響。 |
英文摘要 | The physical transport mechanisms of gas-liquid flows are innately complex and generally entail a great effort to comprehend the nature of the flow field through either experimental measurements or theoretical simulations. Nevertheless, instant knowledge of bubbly-flow characteristics is needed practically for many industrial applications. In this study, an approach for using neural networks is implemented to demonstrate their effectiveness in the real-time determination of fully developed two-phase flow properties of upward bubbly-pipe flows. Three back-propagation neural networks are established via a training process with Liu’s experimental database to predict the distributions of a void fraction and axial liquid/gas velocities of upward two-phase turbulent bubbly flows. Comparisons of the predictions with the test target vectors indicate that the average root-mean-squared errors from three back-propagation neural networks are well within 4.33%. This study also examines the effects of various network parameters, including the number of hidden nodes, transfer function type, number of training pairs, learning rate-increasing ratio, learning rate-decreasing ratio, and momentum value on the performance of neural networks in detail. |
本系統之摘要資訊系依該期刊論文摘要之資訊為主。