請用此 Handle URI 來引用此文件: http://hdl.handle.net/11455/88729
標題: Parameter Optimization for NETSTARS
NETSTARS 模式參數最佳化之研究
作者: Hsieh Hui-Ming
Yang Yin-Lin
謝慧民
楊音琳
關鍵字: back-propagation neural network
network of stream tube model for alluvial river simulation (NETSTARS)
parameter optimization
倒傳遞類神經網路
NETSTARS 模式
參數最佳化
摘要: 本研究以離散參數試誤法、人工經驗調整法與倒傳遞類神經網路法,優選NETSTARS 模式參數,並評估其成效。所用參數為河道曼寧n 值及可沖刷厚度參數Alt 值,模擬對應的成果分別為水位歷程及河床縱斷面高程變化。第一法的推估成果被當成近似理論解,做為評估標準。由水位變動成果發現,以倒傳遞類神經網路法與離散參數試誤法得到的曼寧n 值較為一致,平均值也與人工經驗調整法接近,均適用於NETSTARS 模式,但不同事件所得之最佳參數值仍有些許差異;在河床變動成果的部分,後兩法成果均與第一法差異頗大,由於Alt 值無法由最佳化方法獲得相近的參數成果,因此這些最佳化方法均不適用於此參數之推估。
Methods, such as the experience-based artificial adjustment, back-propagation neural network, and discrete-parameter trial-and-error, were used to investigate the optimal performances of these parameters of the NETSTARS model. The parameters adopted in this study include Manning's n value of channels and Alt value of scouring thickness, and the corresponding results are water level hydrograph and longitudinal riverbed profile. The results of the discrete-parameter trial-and-error method are regarded as approximate theoretical solutions, and it is regarded as a evaluation criteria. The simulation regarding water level change reveals that the Manning's n values resulting from the back-propagation neural network and the discrete-parameter trial-and-error method show the consistency, and the average of these values is also close to the result of experience-based artificial adjustment method. So, those optimization methods are suitable to NETSTRAS model for Manning's n estimation, but the optimized parameters show the significant discrepancy in different events. For the riverbed change, the results of the last two methods vary considerably with the result of the first method. Because similar Alt values cannot be obtained by the optimization methods, these methods are not applicable to this parameter's estimation.
URI: http://hdl.handle.net/11455/88729
顯示於類別:第45卷 第02期

文件中的檔案:
檔案 描述 大小格式 
84621-3.pdf679.47 kBAdobe PDF檢視/開啟


在 DSpace 系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。