台灣地區豐枯季節差異大，降雨多半集中在颱風季節，且因地形崎嶇山高地狹，河短流急無法有效地存蓄雨水，為達到消減洪災及蓄水目的，水庫操作即扮演著關鍵的角色。水庫防洪操作過程為一錯綜複雜的問題，必須掌握水庫上游的即時雨量及水文資訊，加上人類仔細的研判及豐富的操作經驗，方能達到較理想的操作，但操作經驗的傳承卻為一難題。因此，本研究分別根據水庫防洪運轉規則及最佳防洪操作策略建立智慧型水庫防洪操作模式，期能降低下游地區的洪峰量、延長洪峰稽延時間，並在洪峰發生後則調節水庫水位標高以達規線上限位置、確保蓄水功用，以為颱洪期間提供理想防洪操作模式。 本研究於石門水庫建置兩種水庫防洪操作模式：模式1是防洪運轉規則模糊控制模式(FORFCM)，主要根據石門水庫溢洪道閘門操作關係曲線圖轉換成防洪操作之模糊規則，以建置模糊控制模式；模式2為智慧型水庫防洪模糊控制模式(IFOFCM)，是以遺傳演算法搜尋出歷年颱洪事件水庫最佳防洪操作歷程，再以聚類分析萃取出模糊規則以建立調適性網路模糊推論系統(ANFIS)模式。兩模式皆遵循著水庫溢洪道閘門操作之限制來修正放流量，以符合閘門操作的法令規範。並以歷史事件26場颱風以總入流量加以分類後，比較水庫原防洪操作與兩模式操作之洩放歷程；結果顯示兩種模式的防洪操作皆能有效地消減洪峰與達成蓄水之目的，可作為水庫人員參考的重要指標於颱洪時期即時操作模式，並提供其他水庫建置智慧型防洪操作模式之參考。 In Taiwan, the subtropical climate brings rainfall concentrated in the typhoon and plum rain periods from May through October, and the steep mountainous landform makes most of the rainfall flow immediately into the ocean; thus, the rain cannot be directly used as a stable source of water supply. In order to decrease the flood peak stage downstream and store floodwaters for future usage during typhoon seasons, a reservoir flood operation plays a critical role. Real-time reservoir flood operation is a continuous and instant decision-making process based on relevant operating rules, policy and water laws, in addition the immediate rainfall and the hydrology information; however, it is difficult to learn the intelligent experience from the elder operators. The main purpose of this study is to establish the automatic reservoir flood control model to achieve the goal of a reservoir operation during flood periods. In this study, we propose two kinds of flood control models of the Shimen reservoir. These models include two major processes: the knowledge acquired and implemented, and the fuzzy inference system. Model 1 is the flood operation rule fuzzy control model (FORFCM), which is constructed by extracting fuzzy rules from the spillway gate operation relations curves. Model 2 is the Intelligent Flood Operation Fuzzy Control Model (IFOFCM), which is built by extracting fuzzy rules from the rational hydrograph obtained by Genetic Algorithm (GA). Both models comply with the Shimen reservoir''s operating rules and regulations. The results demonstrate that both fuzzy control models can perform much better than the original reservoir operator in 26 flood events and effectively achieve decreasing peak flood stage downstream and storing floodwaters for future usage.