進階搜尋


 
系統識別號 U0026-0812200911395467
論文名稱(中文) 自組非線性系統應用於濁度預測
論文名稱(英文) Study on Turbidity Forecasting by Using Self-Organization Algorithm Model
校院名稱 成功大學
系所名稱(中) 水利及海洋工程學系專班
系所名稱(英) Department of Hydraulics & Ocean Engineering (on the job class)
學年度 93
學期 2
出版年 94
研究生(中文) 王英銘
研究生(英文) Ying-Ming Wang
電子信箱 ingming@mail.water.gov.tw
學號 n8791109
學位類別 碩士
語文別 中文
論文頁數 67頁
口試委員 口試委員-謝啟男
指導教授-顏沛華
口試委員-胡南澤
中文關鍵字 自組非線性  濁度 
英文關鍵字 Turbidity  GMDH 
學科別分類
中文摘要   台灣地區平均年降雨量約2515mm,在空間與時間上分佈極為不均,加上地勢陡峻,河川流短,地質薄弱,沖蝕嚴重,可利用的地表水資源相當匱乏,必須依靠水庫攔蓄水源,使水資源能有效利用。在公共給水中,水庫水源佔相當大比例,而水庫集水區之暴雨易造成土壤沖蝕而流入水庫,土壤中之泥砂、有機物及礦物質導致濁度增加而影響水質。濁度為水質良好與否之重要指標,濁度高的原水在公共給水之處理程序上造成相當之困擾,甚至無法供水,如85年之賀伯颱風、89年之納莉颱風、90年之桃芝颱風、93年之艾莉颱風等,皆造成嚴重之停水事件!因此在自來水之前置處理上,為防止高濁度之進水,主要在於加強集水區水土保持、水源保護區管制,以及引水前濁度之預測與淨水場引水操作時機之掌握。
  
  本研究以自組性(self-organization)網路架構GMDH(Group Method of Data Handling)演算法為基本架構,即以輸入~輸出的觀測資料來建立「濁度預測模式」。其結構建立之觀念源自動物進化或植物配種改良之優選過程。系統之輸入由變數之組合、競爭優選,層層推展至最後誤差收斂或誤差不再改善為止,再由最終輸出層之最佳輸出單元層層回溯至最初之輸入單元,GMDH演算法自行架構衍生組合之最佳高階非線性之輸入~輸出網路架構(模式)即自動形成。

  本文以旗山溪甲仙攔河堰為研究案例,南化水庫水源之一來自該攔河堰之越域引水,因此甲仙攔河堰進水濁度直接影響供水品質,而高濁度的引水將導致水庫淤積造成庫容減少,故本文旨在研究越域引水前的濁度和旗山溪各時期暴雨量、流量和濁度之相關,提供予甲仙攔河堰作引水參考。本研究以2000年5月~2004年12月甲仙攔河堰及其上游每日之水文資料,利用GMDH理論所特有之自組多層演算方法,建立一輸入~輸出關係之「濁度預測模式」進行濁度預測,在這些水文資料中,選定甲仙雨量站之暴雨量、旗山溪總入流量及取水口之濁度作為預測模式之輸入變數建立預測模式,並以遞迴結構GMDH修正原模式,使模式具時變性而能自我調整,以達長期觀測、精確預測的目標。

  據建模結果顯示,本案例以GMDH演算法所預測之結果較佳,且以70筆(日)資料為最佳之建模長度,其中以2003年之效果最為突出。另外本研究經交叉驗證後,選取效果最好的2002年數據之建模作為往後甲仙攔河堰整體的濁度預測模式,其濁度預測成果係以自來水公司一般淨水場處理設備在原水濁度為500 NTU時尚可依設計出水量出水之一成(50 NTU)來當濁度預測精度的指標(90%信賴區間),據測試結果觀之,絕大部分預測值皆位於信賴區間內,表示本研究以自組性網路演算法為架構建立之「濁度預測模式」有良好之實用性。另本研究也嘗試將即時觀測的新資料再納入模式中重新建模,利用遞迴結構對模式參數作即時修正以利隨時掌握最新趨勢,一旦發現模式預測值超越某一門檻值( 50 NTU)而不適用時,即可隨時由實測之新數據進行GMDH遞迴結構之演算來更新模式,經試驗結果顯示,其預測結果亦在信賴區間內,證明本模式亦可適用於濁度之時變性預測。另本研究在甲仙攔河堰之引水操作上,當沉砂池前濁度500 NTU時(原水濁度為500 NTU時尚可依設計出水量出水),流量為180 cms,含砂量為1850 mg/l,故理論上流量180 cms(含)以下為適當之引水時機。但為保守計,以旗山溪含砂量低於1000 mg/l時之流量(約100 cms,甲仙攔河堰沉砂池前濁度約250 NTU)視為引水時機之臨界點。因本研究之濁度預測模式可提供1~10天後之濁度預測,當預測原水濁度為250~500NTU或以下時,該濁度範圍內對應之時段皆宜引水,故本研究之濁度預測模式不失為提供引水操作之實用工具。
英文摘要  Even if average year rainfall is about 2515 mm in Taiwan, the usage of surface water resources is limited due to the cliffy terrain and weak geological structure, steep slope and short stream length, unequal rainfall distribution on space and time. So, reservoir storage becomes a reliable and effective function in most of public water supply. Serious storm can cause soil erosion in catchments and flows into reservoir. Turbidity is then increased and influences the water quality. Turbidity is the most important index for public water supply. High turbidity inflow causes harassment on processing of public water supply even need to cut off the water supply. For example, typhoon Herb (1996), Nari (2000), Toraji (2001) and Aere (2004) all caused severe water non-providing events. In order to avoid high turbidity water inflow, it is important to strengthen the catchment’s conservation, protect the water resources territory, and predict the inflow turbidity concentration before the treatment operation.

 The GMDH (Group Method of Data Handling) algorithm of self-organization network is being used as basic configure in this paper to build up the turbidity predicting model with simply input/output observation data. The GMDH structure originated from the animal or plant evolution process in nature. The optimum high level nonlinear input/output network structure can be self organized by the procedure of variables combination in first layer then compete each other within multiple layers to constrain the error or the error converge no longer, then feed back to its original first layer combination unit to pick up those optimization variables.
 
 Stream inflow of Chia-Shian Weir in Chisan Chi and the relative hydrologic parameters has been provided as study sources for the turbidity prediction. Part of water resource of Nanhua reservoir comes from Chia-Shian Weir through a diversion tunnel. The inflow turbidity directly influences the water supply quality and probably reduces the capacity of Nanhua reservoir. The inflow turbidity of Chia-Shian Weir intake diverted prior into Nanhua reservoir and rain-storm quantity as well the discharge in Chisan Chi are most concerned for inflow diverted operation of Chia-Shian Weir. The daily hydrology data from May of 2000 to Dec. of 2004 such as rainfall of Chia-Shian station, discharge of Chisan Chi and turbidity of Chia-Shian Weir intake were used as the input of this GMDH turbidity prediction model. Then, use the GMDH algorithm to organize variables and result in the best (i.e., minimum estimation error) final relation to build up an optimum turbidity prediction model. The GMDH turbidity prediction model proposed in this paper has regressive mode also. The system can assess the estimate error if over the threshold and self-adjusted the original model by update the field input data which make the model possible to achieve long period prediction and accurate estimation.

 The model construction procedure shows that GMDH algorithm is better than SGMDH (Stepwise regression GMDH) and data length of 70 is the best choice for model construction, among them the result of 2003 is outstanding. Having multiple tests, the best model construction by 2002 data is being selected as the final turbidity estimate model for Chia-Shian Weir. According to the treatment operation procedure of Taiwan Water Corporation, turbidity less than 500 NTU still can be accept to provide water supply normally. The 10% error (50 NTU) is selected to be the index of inflow turbidity prediction (90% confidence interval) and result show that most of estimate values are all within the confidence interval which indicate this turbidity prediction model with GMDH algorithm for turbidity estimation is useful. Also, this model updates the input data from the field observation while the estimate value is over the threshold of ±50 NTU to reset up the model by using the regressive structure for best prediction and tests show that the estimate value are within the confidence interval as well. The critical divert operation is another focal point in this paper. The divert operation opportunity should be in the period of concentration of 1850 mg/l and discharge about 180 cms in Chisan Chi corresponding 500 NTU turbidity in Chia-Shian Weir theoretically but the concentration of 1000 mg/l and discharge about 100 cms in Chisan Chi corresponding 250 NTU turbidity in Chia-Shian Weir be suggested as the critical divert operation for conservative consideration. Because of the 1~10 days prior predicting function, 250~500NTU turbidity can be easily estimated by this remarkable model so as to provide the corresponding critical divert operation period which made this GMDH turbidity prediction model be a practical tool for critical inflow diverting operation.
論文目次 摘 要...............................................................................................................Ⅰ
英文摘要....................................................................................................... III
目 錄................................................................................................................ V
表目錄........................................………………………................................VII
圖目錄........................................................................................................... VIII
照片目錄.......................................................................................................X
第一章 緒 論...................................................................................................1
1-1 研究動機….......................................................................................1
1-2 研究方法..........................................................................................2
1-3 相關文獻回顧..............................................................................3
1-4 本文組織................................................................................4
第二章 GMDH理論........................................................................................6
2-1 基本架構.........................................................................................6
2-2 基本理論.............................................................................7
2-2-1 GMDH演算法.............................................................................7
2-2-2 逐步迴歸之GMDH演算方法(SGMDH)........................................11
2-2-3遞迴結構之GMDH演算方法.......................................................13
第三章 濁度預測模式之建立與驗證...........................................................17
3-1濁度在飲用水的重要性..................................................................17
3-2濁度預測模式之建立.........................................................................18
3-3濁度預測效能評鑑指標....................................................................20
3-4濁度預測模式之驗證(實例驗證)..................................................21
3-4-1南化水庫及甲仙攔河堰概述...........................................................21
3-4-2濁度預測模式參數之選定...............................................................27
3-4-3取水口濁度模式之建立與預測….............................................28
3-4-4資料更新重新建模之預測..............................................................32
第四章 分析結果與討論.......................................................................34
4-1 GMDH及SGMDH之模擬結果與比較…………………................34
4-2輸入變數對模式影響之探討........................................................39
4-3濁度預測結果分析....................................................................45
4-4以新數據重新建模之預測結果比較............................................55
4-5引水方式探討....................................................................59
第五章 結論與建議.....................................................................................62
5-1 結 論..................................................................................................62
5-2 建 議..................................................................................................64
參考文獻…………………………………………………………………..66
參考文獻 1. 顏月珠,「現在統計學」,三民書局。
2. 王如意、易任,「應用水文學」,新編上下冊,國立編譯館出版。
3. 張斐章、黃源義、徐國麟,「自組性演算法及其應用於降雨-逕流模式建立之研究」,台大農學院研究報告,pp.293-316,1994。
4. 張斐章、梁晉銘,「自組性演算法於河川水位流量預測之研究」,台灣水利,第41卷,第四期,pp.14-24,1993。
5. 李梅芳,「自組非線性系統應用於水位相關之研究」,碩士論文,國立成功大學水利暨海洋工程研究所,1993。
6. 楊永祺,「自組非線性系統應用於水位即時預測」,碩士論文,國立成功大學水利暨海洋工程研究所,1997。
7. 周益達,「自組非線性理論建立水位預測模式之探討」,碩士論文,國立成功大學水利暨海洋工程研究所,1998。
8. 李文揚,「自組非線性系統應用於水庫優氧化之預測」,碩士論文,國立成功大學水利暨海洋工程研究所,2000。
9. 吳俊逸,「自組非線性系統應用於地層下陷之預測」,碩士論文,國立成功大學水利暨海洋工程研究所,2001。
10. 財團法人曹工農業水利研究發展基金會,「旗山溪中油段採砂對高雄農 田水利會取水設施影響之研究」。
11. 經濟部水利署南區水資源局,「甲仙攔河堰沉砂池及魚道功能評估研究計畫(沉砂池效率評估部分)」,民國91年7月。
12. 經濟部水利署南區水資源局,「甲仙攔河堰引水水質變化因素及引水時機研究分析計畫」,民國92年3月。
13. Ikeda S.,S. Fugishige,and Y. Sawaragi,「Nonlinear Prediction Model of River Flow by Self-Organization Method」,Int. J. Syst. Sci.,Vol.7,No.2,pp.165-176,1976。
14. Stanley J. Farlow,「Self-organizing Methods In Model-GMDH TYPE Algorithms」,MARCEL DEKKER.INC。
15. Duffy ,J.J. and M. A. Franklin,「A Learning Identification Algorithm and Its Application to an Environmental Systems」,IEEE Trans. System,Man,Cybernetic,Vol.SMC-5,No.2,pp.226-239,1975。
論文全文使用權限
  • 同意授權校內瀏覽/列印電子全文服務,於2006-08-16起公開。
  • 同意授權校外瀏覽/列印電子全文服務,於2006-08-16起公開。


  • 如您有疑問,請聯絡圖書館
    聯絡電話:(06)2757575#65773
    聯絡E-mail:etds@email.ncku.edu.tw