The Study on Real-time Error Correction on Flow Forecasting with Support Vector Machine and Fuzzy Inference Model

• Main Authors: Ming-xaun Li, 李明軒
• Other Authors: Pao-Shan Yu
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/08920035443725638822

碩士 === 國立成功大學 === 水利及海洋工程學系碩博士班 === 96 === This study developed a spatial flow forecasting model by integration of the QPESUMS (Quantitative Precipitation Estimation Segregation Using Multiple Sensor) system with distributed rainfall-runoff model to provide 1-3 hours ahead flow forecasts. Two real-time error correction models were also included to improve the performance of flow forecasting. The up-stream of Wu-Tu flow gauge in Kee-Lung River is used as study area. Ten historical storms occurred during 2006~2007 are chosen as the data bases, in which seven storm events are used for model calibration and another three events are used for model verification. The Micro-Genetic Algorithm (μGA) is utilized for automatic parameters calibration and the searching domain is reduced by the physical property of basin. Furthermore, the study revealed that there are no significant differences between parallel connections calibration (calibrated respectively) and series connections calibration (connected all events and calibrated). In real-time updating, the fuzzy inference method and support vector machine (SVM) are applied to modify the flow forecasts in real time. The results reveal that the integration of real-time error correcting model and spatial runoff forecasting model can improve flow forecasting, however the accuracy still decrease with increase of lead times. The results also showed that SVM method has better performance than fuzzy inference method in most criteria. It seems that SVM is suitable to construct the real-time error correcting model and is able to improve the accuracy of flow forecasting.