翡翠與石門水庫為台灣北部地區供水最重要兩大水庫，不僅擁有各自供水區外，並同時擁有一個共同供水區，其平日操作通常以本身運轉規線（rule curve）為參考依據，然而規線訂立時並未考量鄰近水庫之水資源調度。故本研究運用Non-Dominated Sorting Genetic Algorithm（NSGA-II）搜尋雙水庫系統之聯合供水策略。兩水庫缺水指標（SI）最小為前提，依據各自運轉規線供應個別供水區；並以聯合運轉規線供應共同供水區。 建置平日供水模擬模式必須考量兩水庫之水位、運轉規線、需水量與計算缺水指標（SI），以獲得兩水庫系統之放水量；並運用NSGA-II搜尋法，考量兩水庫硬體及相關操作要點納入限制式，以翡翠及石門水庫缺水指標（SI）為雙目標，並建置三種方案進行討論與比較：方案一為無任何水位限制，試圖搜尋出最符合兩水源系統使用之聯合供水運轉規線；方案二根據翡翠水庫各旬水位變化趨勢加以制定限制式，以求得符合實際現況之平滑規線；方案三更進一步針對防汛期間，考量汛期前後之差異，以提高水資源運用之效率。 歷時49年旬流量資料模擬出之結果顯示，翡翠水庫較石門水庫有能力支援共同供水區，倘若不考慮真實情況與實際操作人員方便性，方案一可得到最佳之pareto解，而方案二與方案三結果非常相近，反之，則以方案三表現最佳。結果指出，NSGA-II可提供多樣化策略於兩水庫系統之聯合操作，而應用於真實系統中，其均勻散佈之pareto最佳解可使決策者易於從眾多非優勢解中得到最適解。 The Feitsui and Shihmen reservoirs are two most important reservoirs for supplying water in Northern Taiwan. The daily operation of the two reservoirs is based on their own rule curves; however, the rule curves are not designed for supporting another reservoir in water supply. They have their own water districts and one common water district. This study applies the non-dominated sorting genetic algorithm (NSGA-II) to find the strategies for the joint operation of the two-reservoir system. They use their own operating rule curves to supply for their own districts and one joint operating rule curve for the common district; besides, they have to assure their own water districts of the minimal water shortage index (SI). A daily operational simulation model is developed to guide the releases of the reservoir system according to the current storage levels, the rule curves, and water demands and then to calculate the shortage indices (SI) of both reservoirs. The NSGA-II is adopted for minimizing the SI values through searching the optimal joint operating strategies. Three alternatives are designed to investigate the joint operation of the two-reservoir system. Alternative 1 does not any constraints for the joint operating rule curves; alternative 2 limits the shape of the joint operating rule curves similar to the shape of the original rule curves; alternative 3 limits the water level of rule curves during flood season in addition to the constraints of alternative 2. Based on 49 years historical data set, the results of these three alternatives show the Feitsui reservoir has water surplus to support the common district. Alternative 1 can obtain the best pareto-front solutions; the pareto-fronts of alternatives 2 and 3 are very close. The results indicate that NSGA-II is a promising approach to offer a number of diversified alternative strategies for the joint operation of the two-reservoir system. The wide spread of pareto-front (optimal) solutions allow the decision maker to facilely determine the best compromise solutions through the trade-off between two reservoirs’ operational strategies and make the method very attractive for real-world applications.