地面地下水聯合運用優選與模擬模式建立之研究

Development of Optimization and Simulation Models for Conjunctive Use of Surface Water and Groundwater

林尉濤(Wei-Taw Lin)

指導教授 ： 徐年盛

國立臺灣大學/工學院/土木工程學研究所/博士(2009年)

https://doi.org/10.6342/NTU.2009.01996

摘要

水資源如何有效利用為當前全球各國家及地區所關切之議題，台灣自1970年代以後經濟蓬勃發展，民生及工業用水需求量急速上升，乾旱缺水現象頻率增加，各標的用水間之水量分配紛爭問題隨之逐年加劇，而農業用水佔總用水量之七成，中部地區更接近九成，近年來，中部地區因工業開發致供水壓力顯著增加，其中，農業用水是否具備調用支援之能力，值得進一步分析探討，本文以優選與模擬模式理論探討地面地下水聯合運用，並以該優選之地面地下水歷線，輔以類神經網路預測未來地下水位之變化，有助於提升水資源利用效益及提供該區域未來發展規劃之參考。
本研究之模式以台灣地區地面水資源及地下水蘊藏量最為豐富之濁水溪沖積扇為案例，經蒐集該區域以往47年長度之水文記錄，進行模式演算，獲得各種不同水文年情形下最佳之地面地下水聯合運用方案，過程中以有無「考量農田水利會轄區外抽水量」，及有無「考量未來中科工業用水量」所組合之4種模擬情境分析，結果顯示，無轄區外及無中科用水之灌溉用水年缺水率SR為1.48%最低，其年缺水指數SI為0.1446，旬缺水指數SI為0.6003，就農業而言可視為無缺水疑慮；而在加入轄區外及中科用水後之灌溉用水年缺水率SR為14.97%最高，其年缺水指數SI為3.3913，旬缺水指數SI為9.0768，顯示已造成缺水之影響；此外，模式中所規劃之工業用水相較於農業用水比例不大，惟在彰化地區地面或地下水資源供應不足之情形下，其缺水指數卻有明顯增加情形，旬缺水指數SI由3.8308增加至23.0245。
此外，運用類神經網路預測濁水溪沖積扇之地下水位，在建立模式之過程中，比較倒傳遞類類神經網路(BPNN)、輻狀基底函數類神經網路(RBFNN)與回饋式類神經網路(RNN)運用在地下水水位預報之精準度，研究結果顯示，類神經網路可推估彰化及雲林地區七個地下水位站在未來一個月內之水位變化，以輻狀基底函數類神經網路表現最佳，倒傳遞類神經網路表現次之，整體而言，在預測30天內之地下水位，模式預測與實際觀測誤差在40公分之合理範圍內，對於該地區地面地下水資源聯合運用、地下水位監控及地下水利用管制等議題具有參考之價值。

關鍵字

水資源利用 ； 地面地下水聯合運用 ； 優選模式 ； 模擬模式 ； 類神經網路

並列摘要

Efficient utilization of water resources is one of the most concerned issues for countries and regions worldwide. In Taiwan due to soaring economy growth after 1970s, the water consumption by domestic and industrial sectors increased dramatically. As the frequency of water shortage increased, the rational allocation of water among utilization sectors becomes a disputed problem year after year. The agriculture consumes nearly 70% of total water consumption, and even approaching 90% in central Taiwan. There has been a significant pressure of water demand from industry development over the past few years in central region. Accordingly, the capability of transferring agricultural water to support other sectors worth further examination. This paper explores the conjunctive use of surface water and groundwater by means of simulation-optimization model theory, and in accordance with the optimal hydrograph of surface water and groundwater applying Artificial Neural Networks to predict future variation of groundwater level. It is expected to raise advantage of water resource utilization and serve as reference for development planning in that region.
The study case focused to the alluvial fan area of Cho-Shui Creek（Choshuishi Alluvial Fan, Taiwan）where kept most affluent reserve of surface water and groundwater. Diverse schemes of conjunctive use were based on hydrological records of last 47 years. Four simulation scenarios were composed that considering the inclusion and exclusion of the pump amount outside the irrigation area of Irrigation Association and the future industrial volume of Central Taiwan Science Park. The results pointed out that, with exclusion of pump outside the irrigation area of Irrigation Association and exclusion of future industrial use of Central Taiwan Science Park, demonstrates the lowest annual irrigation shortage rate (SR) of 1.48%, with shortage index (SI) 0.1446 per year and 0.6003 per 10-day period respectively, which is not a water scarcity problem in regard to agriculture sector. However, the scenario with inclusion of pump discharge outside the irrigation area of Irrigation Association and inclusion of industrial water use of Central Taiwan Science Park would reach a highest SR of 14.97%, and a SI of 3.3913 per year and 9.0768 per 10-day period respectively, which showed the existence of water deficit situation. Furthermore, in model-analyzing and formulation, industrial water took a less proportion than agricultural water. Under water shortage of surface water or groundwater in Changhua district, the water shortage situation increased significantly, with 10-day SI increases from 3.8308 to 23.0245.
Prediction of groundwater level in Alluvial Fan area of Cho-Shui Creek has been applied the Artificial Neural Network algorithms, including BPNN, RBFNN and RNN. The research result revealed that Artificial Neural Networks could give estimation of water table variation in the coming month at seven groundwater monitoring stations in Changhua and Yunlin district, and the RBFNN and BPNN algorithms showed the best performances. The prediction of groundwater level in 30 days, the 40 cm difference was within reasonable range. Findings of the research provide valuable source of information of conjunctive use of surface water and groundwater, groundwater level monitoring, utilization and control of groundwater.

並列關鍵字

Water resource utilization ； Conjunctive use of surface water and groundwater ； Optimization model ； Simulation model ； Artificial Neural Networks

參考文獻

34. 陳俊廷，「應用類神經網路於地面地下水聯合運用規劃之研究」，碩士論文，國立臺灣大學土木工程學研究所，2008。

27. 張良正、何智超，「考量農業用水之區域性地表地下水聯合調配」，國立交通大學土木工程學系，行政院農委會，2004。

29. 鄭文明，「埤塘水源最佳調配調配運用模式之建立與應用」，碩士論文，國立臺灣大學土木工程學研究所，2005。

3. Daliakopoulos, I.N., Coulibaly, P., and Tsanis, I.K.,“Groundwater Level Forecasting Using Artificial Neural Networks,” Journal of Hydrology (2005)309, pp. 229-240 (2005).

5. Karamouz, M., Tabari, M., and Kerachian, R.,“Conjunctive Use of Surface and Groundwater Resources: Application of Genetic Algorithms and Neural Networks,” World Water Congress (2004).