新型跨區水資源調度系統之研究

台灣近年受氣候變遷影響乾旱日趨顯著，台灣年總降雨量雖然豐沛，但降雨時間分布不均，尤以台灣南部地區之河川流量豐枯比約為9:1最為嚴重，此外南部地區水資源系統水源調配不均的問題，使區域水資源調度壓力更形加劇，水資源短缺發生的頻率和幅度將在未來逐漸增加，因此未來長期入流量的推估、水庫最佳化操作和水資源跨區整合調度為水資源有效利用重要且迫切之課題。台灣南部地區水資源互相支援規則複雜，為了水資源應用效益最大化，急需一個系統跨區整合調度。爰此，本研究提出一跨區水資源調度系統，採用人工智慧演算法，建構跨區水資源調度系統，包含區域供水比例推估以及多目標最佳化水庫操作策略制定。其中，供水比例推估模式建置採用支持向量機、多層感知器、深度神經網路、隨機森林演算法、長短期記憶、門閘遞迴單元共六種模式，使用評鑑指標及透過網格搜尋法尋找最佳的模式及其使用的因子與參數，回歸分析各因子與入流量的非線性關係，以學習區域水庫之供水行為，整合區域內水庫之蓄水量，並結合近年新穎的啟發式演算法之一，非支配排序遺傳演算法(Non-dominated Sorting Genetic Algorithm, NSGA-III)，訂定水庫農業供水量最佳化規則，以制定包括公共、工業和農業用水及期末蓄水量等多目標最佳化水庫操作策略，應用長期入流量推估，提供未來六個月水資源逐日操作建議。本研究以台灣南部地區嘉義、台南以及高雄水資源系統驗證模式之準確度。首先解析南區水資源的調度原則，在既有原則下以NSGA-III和人工智慧理論，提出最佳化操作策略，並將水資源調度系統調控之成果與歷史數據對比分析。結果顯示，區域供水比例推估模式以SVM表現為最佳，能準確模擬台南地區實際供水行為。以2014枯水年嘉南灌區調度成果為例，根據NSGA-III演算結果制定之最佳化操作策略，可改善當年歷史灌溉供水量不足9149.5萬噸狀況，且動態調控期末供水量較歷史紀錄增加約6448.3萬噸，供水量提升51.4%，期末蓄水量亦可蓄留7122.7萬噸，達蓄水量管控目標。本研究提出之跨區水資源調度系統，可提供未來六個月跨區和跨標的之水資源調度建議，並提前示警水資源缺乏，建議農業停灌，透過調度減少缺水導致的不便，面對未來異常氣候之挑戰，期待提高水資源使用效益，並提供相關單位作為水資源調度決策的參考。

關鍵字

水資源調度系統；跨區調度；人工智慧；非支配排序遺傳演算法；最佳化操作

並列摘要

In recent years, the influence of climate-related issues such as climate change and droughts on Taiwan has become increasingly significant. Taiwan has been suffering from the non-uniform distribution of precipitation in time, despite the total annual precipitation is abundant. The issue is serious particularly in southern Taiwan which the flow in the wet season accounts for 90% of the total flow. Furthermore, the imbalanced distribution of water resources in southern Taiwan leads to difficulties in allocating water resources. Since the frequency and intensity of water shortage will gradually increase, the long-term simulation in inflow, reservoir optimization, and allocation of cross-region water resources are necessary for effective water resources utilization. The regulations of supportive water resource allocation in southern Taiwan are complicated. In order to maximize the benefit of regional water allocation, this study presents a cross-region water resources allocation system that constructed by artificial intelligence. The system can not only simulate the water supply strategies of multi-reservoir, but also develop the multi-objective optimization operations to trade-off between the domestic, industry, agricultural water demands, and reservoir storage. Artificial intelligence methods/algorithms used in this study to construct simulation model include support vector machine (SVM), multilayer perceptron, deep neural network, random forest, long short-term memory, and gates recurrent unit. Gridsearch is used to calibrate the factors and parameters of the models, and the best model is determined according to the performance measures. Then, the cross-region water resources allocation system integrates the best model mentioned above and Non-dominated Sorting Genetic Algorithm (NSGA-III) to set optimal agricultural water supply regulations. The system can provide every-day operation recommendations within a six-month period with the long-term streamflow forecasting. In this study, the water resources systems of Chiayi, Tainan, and Kaohsiung are chosen to verify the accuracy of the cross-region water resources allocation system. The results of this study include the following three parts: Firstly, the water resources allocation principles are analyzed. Secondly, an optimal operation strategy based on NSGA-III under the existing limitations is proposed. Finally, we compare the results of water resource allocation system with historical data. The results show that the performance of SVM model is the best. Take the allocation result of Chianan irrigation area in 2014 for instance, the proposed operation can improve the situation in which irrigation water supply was insufficient for about 91.5 megaton. In this case, the suggested water supply from the model is a 64.5 megaton increase from historical record. That is, rise by 51.4%. Simultaneously, the proposed water storage is 31.2 megaton more than the water storage control objective. The allocation system for cross-region water resources can propose operating recommendations within a six-month period to mitigate the inconvenience caused by water shortage, warn the insufficiency of water resources in advance, and propose the suggestions of suspending irrigation. In conclusion, the model proposed in this study can improve the efficiency of water resources utilization and assist local agencies take on the challenges of abnormal climate in the future.

並列關鍵字

water resources allocation system ； cross-regional allocation ； artificial intelligence ； NSGA-III ； optimal operation

參考文獻

1. Liao CY (2021) Long-term streamflow forecasting using an AI-based rainfall-runoff model with TCWB1T1 output. Master's theses, National Taiwan University

Google Scholar

2. John H (1992) Genetic Algorithms. Scientific American 267:66–73

Google Scholar

3. Sharif M, Wardlaw R (2000) Multireservoir Systems Optimization Using Genetic Algorithms: Case Study. Journal of Computing in Civil Engineering 14(4):255–263

Google Scholar

4. Van Zyl J, Savic D, Walters G (2004) Operational Optimization of Water Distribution Systems Using a Hybrid Genetic Algorithm. Water Resources Planning and Management 130:160–170

Google Scholar

5. Feng Z, Niu WJ, Cheng CT (2018) Optimization of hydropower reservoirs operation balancing generation benefit and ecological requirement with parallel multi-objective genetic algorithm. Energy 153:706–718

Google Scholar

主題瀏覽