智慧型水資源調配策略以因應用水需求成長

Abstract

臺灣地區近年來由於人口成長及工商產業發展情況下，用水需求逐年增加，乾旱期間用水不足時，多以停灌休耕之方式解決用水不足之窘境，調用農業用水已穩定供應公共用水需求，面對地區水資源時空分布不均及日益不足等問題，如何在不同用水需求下進行水庫操作使其儘可能滿足各用水標的，以善用水資源並維持其永續經營為當前首要課題。藉此，為確保水資源及農業永續經營與發展目標，本研究擬建置因應用水需求成長之智慧型農業水資源調配策略，提供在乾旱時期最佳調配策略下，稻米一期作期間的季缺水指標推估資訊，以協助於水庫管理人員能即早進行評估與決策作業。 本研究以石門水庫作為研究區域，根據文獻回顧以及水庫歷年多目標運用統計資料整理，探討石門水庫供水目標的轉變，擬訂出九種未來可能發生的用水需求情況，針對不同的需水情況，透過系統分析方法，以模擬法(based on M-5 rules)及優選法(NSGA-II optimal search)對水庫放水序列進行模擬及搜尋，結果以優選法能搜尋到較低的缺水率，證實非支配排序遺傳演算法-II對於水庫操作問題具有良好的表現。 類神經網路具有學習、處理複雜的問題與不確定性的特性，本研究透過建置倒傳遞類神經網路(BPNN)及調適性網路模糊推論系統(ANFIS)，以用於推估乾旱時期季長期缺水率，篩選具支配性的輸入因子及訂定門檻值，結合優選法所搜尋出來的缺水率用於網路的輸出端上進行網路訓練，研究結果顯示類神經網路對於公共及農業缺水率皆具有精準的推估， ANFIS模式在不同子集合設定上皆具有不錯的表現，提供成為水資源調配管理之參考依據。

The population growth and economic development in Taiwan has led to a tremendous demand for natural water resources. In recent years, water shortage problems frequently occur in northern Taiwan such that water is usually transferred from irrigation sectors to public sectors during drought periods. In response to the uneven spatio-temporal distribution of water resources and the problems of increasing water shortages in this region, it is a primary and critical issue to simultaneously satisfy multiple water objectives through adequate reservoir operations for integrated water resources management. For sustaining water resources and agricultural development, this study intends to build up the optimal agricultural water resources allocation strategies adapting to the growing water demands of both agricultural and public sectors. The optimal allocation strategies are expected to adequately suggest quarterly water shortage indexes for the period of the first paddy crop such that early assessment and decisions on water allocation can be made for drought mitigation management. The Shihmen Reservoir in northern Taiwan is used as a case study. According to previous studies and historical multi-objective reservoir operation data of the Shihmen Reservoir, this study investigates the changes in water supply targets and design possible nine water demand conditions that may occur in the future. Based on these designed conditions, we use a system analysis approach to conducting the simulation (based on M-5 rules) and optimization search (by the non-dominated sorting genetic algorithm-II (NSGA-II)) of the reservoir operation sequences. The results indicate that the NSGA-II method can search the optimal water allocation series meeting the objectives subject to restrictions and producing a lower water shortage index. It demonstrates that the NSGA-II produces good performance for reservoir operation problems. Artificial neural networks (ANNs) have the ability to learn and deal with complex problems and uncertainty issues. In this study, the back-propagation neural network (BPNN) and the adaptive network fuzzy inference system (ANFIS) are used to estimate quarterly water shortage indexes in drought periods for agricultural and public sectors based on the selected input factors and the determined thresholds. The shortage index obtained from the NSGA-II is the training target of the output layers for both ANN models. The results indicate that the BPNN and the ANFIS models have equally good performance in estimating the shortage indexes for both sectors, but the ANFIS model produces better performance in stability. This proposed approach can be used as an early warning system for drought mitigation, which can become a reference guideline for sustainable water resources management.