本研究將Adaptive-Network-Based Fuzzy Inference System(ANFIS)模式應用於都市雨水下水道水位站雨水下水道水位預警模式(ANFIS for Sewer Stage Forecasting,簡稱ASF模式),並結合Web GIS實作一雨水下水道水位預警系統(ANFIS for Sewer Stage Forecasting System,簡稱ASFS系統),展示水位站地理位置與水位預警結果。以台北市第四排水分區的中港與玉成抽水站系統作為研究區域,此系統在颱風暴雨期間會進行每分鐘監測水位資料紀錄,因此,本研究利用2004~2006年颱風事件共8場,抽水站水位資料,作為ASF模式建立之輸入資料,以5場颱風事件資料作為訓練,2場颱風暴雨事件資料作為驗證,1場颱風事件做為測試,並探討雨量與抽水站水位的相關性,以及在不同超前預測時距(如提前5、10、60分鐘)下ASF模式預測結果的精確性與實用性。經由初步模擬結果,雨量對於抽水站水位的相關性較低,不適合作為抽水站水位模擬預測之輸入資料,而ASF模式在超前預測5分鐘與10分鐘有較佳之預測效果。
The study applies Adaptive-Network-Based Fuzzy Inference System (ANFIS) to sewer stage forecasting, called as ASF model, with Web GIS to set up a ANFIS for Sewer Stage Forecasting System, called as ASFS. In Taipei city, Zhung-Gung and Yu-Cheng drainage systems are employed for the practical case study in that the drainage system minutely record observed water stages in each storm event. Accordingly, ASF model adopts the observation data from 8 storm events in 2004~2006 as ASF model inputs. Of the 8 storm events, 5 storm events are adopted for the training set; 2 storm events, for validation set; and 1 storm event, for test case. Investigations regard 1) the correlation between rainfalls and water stages, and 2) the prediction accuracy and applicability due to computing time, by ASF model based on the different forecast periods, 5 minutes, 10 minutes and 1 hour respectively. As a result, rainfall data is not proper to be used as the ASF model inputs due to weak correlations between rainfall data and output water stages. In consequence, ASF model provides excellent predictions for 5-minute and 10-minute forecast periods.