驟雨衝擊效應之流量即時預報

Real-time Flood Forecasting by Considering the Rain-burst Effect

萬本立(Been-Lih Wan)

指導教授 ： 鄭克聲

國立臺灣大學/生物資源暨農學院/生物環境系統工程學研究所/碩士(2012年)

https://doi.org/10.6342/NTU.2012.00046

摘要

台灣近年來極端降雨事件發生次數日益頻繁，造成許多的洪水災害，使得居民生命、財產遭受嚴重威脅。為降低洪水災害損失，即時流量預報就成為重要的研究課題。
本研究以曾文水庫集水區和赤蘭溪流域為研究區域，針對研究區域颱風、豪雨事件之流量資料，以時間序列的AR(2)模式和Naïve模式做為流量即時預報模式，並藉由雨量資料來改善時間序列模式預報延遲的問題。雨量資料的探討分為兩部分，第一部份為探討雨量增加量和流量增加量之間的關係，藉由找出會造成流量突增的雨量增加量(rain-burst)，建立雨量增加量和流量增加量之間的關係式，結合AR(2)模式修正驟雨效應所導致的預報延遲誤差。第二部份為應用單位歷線的概念，以線性迴歸方式建立雨量差和流量差的反應函數式，由前幾個時刻的雨量差推求預報時刻的流量差，結合Naïve模式進行即時流量預報，藉由考慮到雨量變化的整個趨勢，大幅改善預報延遲問題。
研究結果顯示，考慮rain-burst的AR(2)模式可藉由降低峰值時刻的預報誤差，提高CP值，改善預報延遲的問題，而結合反應函數式的Naïve模式在CP值的表現，明顯優於其他模式，顯示考量整體雨量變化趨勢對於改善預報延遲問題有很大的成效。

關鍵字

時間序列 ； AR(2)模式 ； Na&iuml ； ve模式 ； 驟雨效應 ； 單位歷線 ； 反應函數

並列摘要

In recent years, the extreme rainfall events become more and more so that result in many flood disasters that make residents’ lives and property suffered a serious threat. In order to reduce flood damage, real-time flood forecasting has become an important research topic.
Research analysis was processed with flood events of Tseng-Wen Reservoir Watershed and Chi-Lan River basin. This study constituted several forecasting models of hourly stream discharge based on AR(2) model and Naïve model, and correct the problem of forecasting time lag phenomenon by considering rainfall data.
The discussion of rainfall data is divided in two parts. First part is that discuss the relationship between increment of rainfall and increment of discharge. By identify the increment of rainfall (rain-burst) which can make discharge significantly increase in a short time, we can establish the function of relationship between increment of rainfall and increment of discharge and combined with AR(2) model to correct the problem of forecasting time lag phenomenon which result from rain-burst effect. Second part is that apply the concept of unit hydrograph to establish response function between rainfall difference and discharge difference by linear regression, use data of rainfall difference before prediction time to estimate discharge difference on prediction time and combined with Naïve model to forecast hourly discharge. By considering the trend of rainfall variations, significantly improve the problem of forecasting time lag phenomenon.
The results of research shows that AR(2) model by considering the rain-burst effect can improve the problem of forecasting time lag phenomenon and enhance CP value by reducing the prediction error on peak time. And the performance of Naïve model which combined with response function is significantly better than other models. This result demonstrates that considering the trend of rainfall variations is very effective to improve the problem of forecasting time lag phenomenon.

並列關鍵字

Time series ； AR(2) model ； Na&iuml ； ve model ； Rain-burst effect ； Unit hydrograph ； Response function

參考文獻

2. 王俊欽(2001)，「基隆河流域洪流時序分析模式」，私立中原大學土木工程研究所碩士論文。

8. 鍾侑達(2007)，「以類神經模糊推理修正水位預測之一階延遲現象」，私立逢甲大學水利工程研究所碩士論文。

9. Wu, C.L., Chau, K.W. (2010) Data-driven models for monthly streamflow time series prediction. Engineering Applications of Artificial Intelligence, Vol.23, pp.1350–1367.

10. Toth , E., Brath, A., and Montanari, A. (2000) Comparison of short-term rainfall prediction models for real-time flood forecasting. Journal of Hydrology , Vol.239, pp. 132–147.

11. Robert, J., Alison, J., and Linda, M. (2007) Timing error correction procedure applied to neural network rainfall–runoff modeling. Hydrological Sciences–Journal–des Sciences Hydrologiques, Vol.52, pp.414-431