桃園農田水利研究發展基金會

Plan Name

Study on Six Hours ahead Storm Forecast for Shihmen Reservoir Watershed

計畫編號

88-2-04

類別

灌溉營運管理類

摘要

水利資訊（Hydroinformatics）或稱水利資訊學，係利用計算機與資訊科技的發展，將各類數值模式與資料結合，應用在水資源環境之模擬，可大大提高決策評估的正確性。隨著電腦硬體與軟體的發展迅速，電腦科技也普遍應用在工程界，類神經網路具有學習及聯想的優點，可以模擬非線性的系統，目前也廣泛應用在很多領域。本研究採用類神經網路配合中央氣象局所發佈之六小時颱風資料、中正機場都卜勒氣象雷達資料，以及石門水庫上游集水區十個雨量站的雨量資料，建立颱風降雨預測模式。結果顯示，類神經網路在各模式的學習過程中，皆可呈現出高度相關，表示其在複雜非線性關係中具有良好之學習能力；然而在驗證過程中，由於雨量站位置與颱風事件間錯綜複雜關係，使得結果呈現時好時差現象；一般雨量站大都欠缺氣象觀測記錄，亦使得模式呈現不佳結果，故氣象觀測資料是否具代表性，將影響降雨預測之準確度；綜觀整個降雨模式，影響因子的決定有待更進一步的探討。颱風降雨的物理機制非常的複雜，類神經網路本身屬於非線性計算系統，具有學習與回想的能力，利用類神經網路模式來建立颱風降雨模式，尋找降雨因子與降雨量之間非線性的關係，對於颱風降雨系統的模擬可以提供一個比較合理方法。雷達具有空間偵測能力，利用地面雨量計的雨量資料與雷達回波的資料，可以將空間中的降雨分佈趨勢表示出來；本研究當中所決定的颱風降雨因子是根據過去相關研究文獻所整理出來，而這些颱風降雨因子是否能完全模擬颱風降雨的機制，尚待觀察。本研究主要針對暴雨進行降雨量預測，提供流域集水區降雨量的推估，配合平均雨量的計算方法，來推估集水區流域的平均降雨量，可作為集水區水文模式建立的基礎，或者是洪水預報決策系統的參考。

Abstract

Due to rapid growth of the computer software and hardware, hydroinformatics, a combination of numerical model development and information technology, becomes a useful tool in hydraulic engineering. Neural network algorithm has the capability of learning and thinking, and thus can be applied to simulate a nonlinear system. Currently is widely applied in various fields.Neural network is adopted in this study to simulate the rainfall intensity during a typhoon event. Data used including the 6-hour typhoon data released by the Central Weather Bureau, Dopplor Radar data of C.K.S international airport and 10 rain gauges in the watershed area of Shiemen Reservoir.High correlations were observed during the learning process of the neural network. It indicates that good learning capabilities exist among these complicated nonlinear relations. However, due to quality of the radar data and lack of some meteorological data during typhoons event, some of the simulation results are not satisfactory. Hence, good quality of the key meteorological data are crucial to the accuracy of the simulation results. Determinations of these key parameters require further investigations. Mechanism of the rainfall process during a typhoon event is very complicated. Using a neural network algorithm to establish a typhoon rainfall model and to seek the nonlinear relations between the relevant parameters and rainfall intensity is a rational approach. The rainfall data used in this study include rain gauge data and radar data, and is able to reflect the spatial distribution of the precipitation pattern. The key parameters that affect the rainfall process were determined through literature survey. If these parameters are able to reflect this complicated rainfall process, needs further investigation. The results of this study will provide a useful tool in rainfall prediction and will serve as a foundation of the establishment of hydrology model in a watershed area.

計畫主持人

李鴻源

Project Director

Hong-Yuan Lee

關鍵字

暴雨預測、石門水庫

Keywords

Storm Forecast、Shihmen Reservoir

成果報告