本研究嘗試結合支撐向量分類(support vector classification, SVC)及倒傳遞神經網路(back-propagation neural network, BPN)兩種模式，建立一個颱風降雨預測模式。SVC 在二元分類以及BPN 在非線性系統模擬上，兩者均有相當優異之表現，因此本研究嘗試結合這兩種模式，將其應用於颱風期間之即時雨量預測。本研究以台北地區1980 年至1998 年之颱風雨量資料為應用例，共選取32 場颱風事件作為模式訓練、驗證及測試，其中24 場作為模式之訓練、6 場作為驗證及2 場颱風作測試。首先，以單一雨量站之雨量資料加上8 個颱風因子作為模式之輸入項，以BPN 進行延時1 至3 小時之驗證，結果顯示延時一小時的成果較好，故選用延時1 小時作為主要輸入項目。接著，以SVC 進行二元分類，發現其平均分類之正確率未達70%，且其驗證及測試雨量資料之預測結果亦不佳。然而，若
假設分類正確率達100%之情況下，則其降雨預測之結果會有大幅改善。建議未來應針對如何提高SVC 分類之正確率加以探討，藉以加
強此模式之預測結果。

The support vector classification (SVC) and back-propagation neural network (BPN) have good performance in binary classification and nonlinear system modeling, respectively. This study attempts to combine these two models in order to forecast the typhoon rainfall. In this study, rainfall records for the 10 rain gauges are available from 1980 to 1998. From these records, a total of 32 typhoon rainfall events are selected. The typhoon characteristics of these 32 events are used to develop the forecasting model. Of these 32 typhoon rainfall events, 24 events are selected as training set, six events are taken as the validation set, and the remaining two events are used as a testing set for testing the forecasting performance of the BPN model. First, lag-1h, lag-2h and lag-3h models are tried to determine an order of input variables. The results show that the lag-1h model is the best for each rain gauge. Hence, in this study the present rainfalls and typhoon characteristics are the influential factors of the one-hour ahead typhoon-rainfall forecasting.
Then, the SVC is applied to binary classification of the rainfall data. It is found that the average correct classification rate is only 70%, and the BPN model performs worse. We assume that if the correct classification rate is 100%, then the BPN model will perform well. Therefore, more studies and discussions will be carried out on raising the accuracy of SVC classification for improving the generalization ability of the model in the future.