應用禁忌演算法進行降雨－淹水警戒值之優化：以臺北市文山區及新北市新店區為例

臺灣飽受季風盛行區域產生的短延時強降雨或是每年平均3~4次颱風暴雨所導致淹水災害造成的財產及人員損失。據此，經濟部水利署採用非工程手段建立全臺灣降雨-淹水預警系統，當觀測累積雨量到達或超過該區域設定的降雨-淹水警戒值時，預警系統會發出提醒警戒區域若降雨持續將有淹水之虞。然而部分警戒區域之警戒值設定不佳，仍有改善之空間。本研究收集臺北市文山區及新北市新店區淹水災害事件及歷年時雨量資料，爾後透過雨場分割法劃分各降雨事件的開始及結束時間點，再利用禁忌演算法針對各降雨事件之1、3、6、12及24小時累積雨量之五種降雨延時同時演算最佳化降雨-淹水警戒組。研究結果顯示，本研究設定之降雨-淹水警戒值具有良好的預警能力，其精度最大可提升40%，淹水時命中淹水結果之準確率最大可提升33.3%，而預警時命中淹水結果之準確率最大可提升5.7%。

關鍵字

降雨-淹水警戒值；累積雨量；災害事件；雨場分割法；禁忌演算法

並列摘要

Taiwan located in Southeast Asia faces strong monsoon and three to four typhoons annually which may cause floods and may result in great damage as well as loss of property. The Water Resource Agency (WRA) has established a warning system as a non-structural measure. The concept of the warning system is based on rainfall threshold. When the cumulative rainfall reaches or exceeds rainfall threshold, the warning system will alert emergency response personnel. According to flood records provided by WRA between 2013 and 2015, the records of flood alarm showed that the warning accuracy of warning system needs to be improved. This study provides a methodology to build effective rainfall thresholds through the tabu search which optimizes rainfall thresholds based on 1、3、6、12 and 24 hour duration at the same time. The cases of study are in the Wenshan and Xindian district. And the rainfall events are divided into four situations, including correct alert (A), missed alert (B), false alert (C) and correct rejection (D). The results show that the rainfall thresholds of study have more alerting ability than official rainfall thresholds. The increasing maximum of accuracy is 40%, the increasing maximum of hitting flood during flooding time is 33.3%, and the increasing maximum of hitting flood during warning time is 5.7%.

並列關鍵字

rainfall threshold ； cumulative rainfall ； flood disaster events ； inter-event time definition ； tabu search

國際替代計量

應用禁忌演算法進行降雨－淹水警戒值之優化：以臺北市文山區及新北市新店區為例

全文下載

主題瀏覽