| Summary: | 碩士 === 輔仁大學 === 應用統計學研究所 === 98 === In Taiwan, the plum rains and typhoons often bring about torrential rains and cause serious floods. If the flood could be predict on time, it would become effective reference and help people to make decision for the river basin's flood response. Moreover, it could reduce demands and social economic. In recent research of river stage forecasting, non-real-time predict method is commonly used which is called artificial neural network (ANN). The drawbacks of ANN are operation time was tediously long, difficult to select input variable and model doesn’t have interpretation. In order to reach the purpose of predict real-time river stage, the method must have robustness to outlier and missing value. MART is a robust method which has fast learning speed, has capability of variable selection and establishes variable importance. However, there is no related research to discuss if this approach could use in the real-time river stage forecasting or not. In this study, using the related rainfall and river stages data of the sixteen typhoon flood events during the 2005 to 2009 years. According five different evaluated index to compare the performance about MART and other non-real-time data driven method which are MARS, SVR, GAM, ANN. The results reveal that the river stage flood forecasting method MART has better efficacy and could provide more accuracy and effective data in the river basin's flood response decision-making.
|
|---|
