調適性網路模糊推論系統於水庫操作之研究

Abstract

面對台灣地區水資源時空分布不均及日益不足等問題，如何在安全條件下進行水庫操作使其儘可能滿足各標的，以善用水資源並維持其永續經營是當前首要課題。智慧型控制理論主要包含人工智慧、模糊理論、遺傳演算法與類神經網路等技術，模擬人類學習、適應、回想等諸多能力，以解決模式不確定系統、非線性與時變系統等傳統方法不易解決之問題。本研究主要目的是利用智慧型控制理論建立不同時期水庫即時操作系統，期能提供水庫管理單位於操作運轉之參考；首先建立水庫平時操作系統，藉由智慧型理論之彈性及傳統規線操作之專家知識相結合，提出智慧型水庫操作策略，以石門水庫過去四十年的資料作為訓練與測試，相較於傳統規線操作之結果，可有效改善缺水情形集中之嚴重乾旱現象；其次探討反傳遞模糊類神經網路(CFNN)與調適性網路模糊推論系統(ANFIS)於水庫入流量預測之應用，結果顯示ANFIS具備高度學習能力，利用少數規則即可精確預測水庫入流量。再以模糊規劃理論建立一颱洪時期水庫優選模式，利用遺傳演算法搜尋颱洪時期之水庫最佳放水量歷程，以茲作為ANFIS之訓練樣本與標的，而水庫颱洪操作系統需要靠迅速而精確地掌握水庫集水區上游雨量及流量站水文狀況，因此利用ANFIS進行水庫入流量推估，進而在安全與減災的前提下，建立一智慧型水庫颱洪操作模式，於洪水期間進行有效的水庫颱洪操作以充分利用大自然所帶來的資源，並提供操作人員進行相關決策之參考數據。

Resulting from the continuous increase in water demand and uneven water distribution both on time and space, the efforts of pursuing integrated optimal water resource management become critical. Intelligent control is a state-of-the-art technology that resembles the human thinking process in decision making and strategy learning, and it has been well recognized for its outstanding ability in controlling complex systems In this study, we continue to pursue the novel intelligent control methodology, which includes genetic algorithm (GA), fuzzy theory, and adaptive network-based fuzzy inference system (ANFIS), for efficiently and effectively operating a multi-objective reservoir. First, GA and fuzzy rule base (FRB) are used to extract the knowledge based on the historical inflow data with a design objective function and the traditional rule curve operating strategy, respectively. The ANFIS is then used to implement the knowledge, to create the fuzzy inference system, and then to estimate the optimal reservoir operation. The results show that the ANFIS models built on different types of knowledge have better performance than the traditional M-5 rule curves in reservoir operation. Second, for the purpose of comparison, the predictive ability of CFNN and ANFIS model are performed by using the inflow data of Shihmen reservoir. The result demonstrate that ANFIS can be applied successfully and provide high accuracy and reliability for reservoir inflow forecasting in the next three hours. Finally, we propose an intelligent operating strategy for real-time reservoir flood operation. The fuzzy programming is implemented to deal with the inherent imprecision and vagueness characteristic of reservoir objectives and constraints; GA is then to search the optimal solutions; forecasted hourly inflow by ANFIS and observed hourly inflow are used to simulate the model performances, respectively; ANFIS is used for estimating the optimal flood operation. The results obtained in this study are valuable information for the flood mitigation.