洪水時期水庫最佳操作之研究—以石門水庫為例

A Study on the Optimum Reservoir Operation during Floods – for Shihmen Reservoir

黃俊傑(Chun-Chieh Huang)

指導教授 ： 王安培

中原大學/工學院/土木工程研究所/碩士(2004年)

https://doi.org/10.6840/cycu200400421

摘要

I
摘要
台灣年平均降雨量雖高達世界平均年降雨量的2.5 倍，但人口密度
高，以致於每人每年平均分配雨水量不足。又台灣的降雨在時空上變化
很大，大部分集中在每年5 到10 月的豐水期，期間又以颱風所帶來
的充沛雨量為主。因此，在颱風所造成之洪水時期，水庫的操作顯得格
外重要，如何在此期間求得最佳水庫放水量，使水庫本體安全、下游不
因水庫放水而使河川水位過高、颱風過後水庫儘可能達到滿水位，實為
解決臺灣水資源困乏的方法之一。
本研究以石門水庫，及其上游集水區、下游新海大橋為研究對象，
並利用民國86 至90 年間颱風侵台時期之資料，首先嘗試以類神經網
路（Artificial Neural Network，簡稱ANN）建立石門水庫集水區降雨-
逕流模式，以推估石門水庫未來6 小時可能發生的入流量，以利防洪
及預警之用；接著以大漢溪下游新海大橋水位站為研究地點，以類神經
網路分析水庫洩洪時對下游水位增量之影響；最後使用遺傳演算法
（Genetic Algorithm，簡稱GA），在安全的範圍內，優選出在洪水時
期水庫之最佳放水量，期望能達到水庫最佳操作，以供石門水庫即時操
作之參考。
在水庫最佳操作模式中，本研究嘗試將一場颱風以兩時間切割點
（A , B）分割為初期、中期與後期，並使用水庫水位限制線，將水庫水
位限制一安全水位之下。模式結果顯示，以時間切割點為18 與42
時，較適合侵台時間為48 小時左右之颱風；而時間切割點為24 與48
時，在侵台時間較長的颱風上有不錯的表現。
關鍵字：類神經網路、遺傳演算法、水庫最佳操作、降雨-逕流模式

關鍵字

遺傳演算法 ； 降雨-逕流模式 ； 水庫最佳操作 ； 類神經網路

並列摘要

II
Abstract
Although the mean annual precipitation in Taiwan is 2.5 times the
average value in the world, there is a shortage of mean precipitation per
person per year due to intensive population. The precipitation is unsteady in
different places and during different seasons, and it usually concentrates
during May to October, the moist season, due to the typhoon that brings in
juicy rain. Therefore, it is important to get the best reservoir drainage to keep
it safe. Reducing higher water-level in downstream due to drainage of
reservoir, and making reservoir to achieve the full level as soon as possible
after typhoon is over can solve the shortage of water resource in Taiwan.
We took Shihmen Reservoir, its watershed and Hsin-hai bridge in
downstream as examples; and the data of invading typhoon in 1997 to 2001
are used in this paper. At first, the Artificial Neural Network (ANN) is used
to simulate the rainfall-runoff model and to predict the inflows of 6 hours
later in Shihmen Reservoir. Secondly, we focus on Hsin-hai Bridge in
Ta-han Stream Basin. The ANN is applied to analyze the water stage
increment of the river due to Shihmen Reservoir drainage. In the last phase,
Genetic Algorithm (GA) is used to select the best drainage of reservoir
during the period of flood and to achieve the optimum reservoir operation. It
is expected that results of this study could be used for online reservoir
operation in the future.
For the optimum reservoir operation model, this study attempts to
divide the process of typhoon into three phases which includes beginning,
middle and ending by two time points A and B. The safety line of reservoir
is utilized to limit the water-level. The results show that the safety line of
reservoir at time points A=18, B=24 and A=24, B=48 corresponded to
48-hour and over 48-hour typhoons respectively are satisfactory.
Keywords: Artificial Neural Network (ANN), Genetic Algorithm (GA),
optimum reservoir operation, rainfall-runoff model

並列關鍵字

optimum reservoir operation ； rainfall-runoff model ； Artificial Neural Network (ANN) ； Genetic Algorithm (GA)

參考文獻

46.張大元（2003），「類神經網路在水庫放流對河川水位增量之研究」，中原大學土木工程研究所碩士論文。

56.梁亨如（2003），「應用遺傳演算法於ASR井操作之研究」，中原大學土木工程研究所碩士論文。

64.楊元鎮（2003），「集水區可能最大洪流量之研究」，中原大學土木工程研究所碩士論文。

1.Amorocho, J. and Hart, W. E., 1964, “A critique of current methods in hydrologic systems investigation,” Transactions of American Geophysical Union, Vol. 45, pp.307-321.

2.Asaad, Y. Shamseldin, Kieran, M. O'Connor and Liang, G. C., 1997, “Methods for combining the outputs of different rainfall–runoff models,” Journal of Hydrology, Vol. 197, pp.203-229.