| Summary: | 碩士 === 國立臺灣大學 === 生物環境系統工程學系暨研究所 === 91 === Flood forecasting is crucial for the safety of reservoir operation systems. Building an accurate and effective streamflow forecasting model has always been a major goal of hydrologists in Taiwan. Various types of artificial neural networks (ANNs) have been used to construct the rainfall-runoff processes. The input layer of ANNs is usually established by using the several preceding hour data of the upstream rainfall and streamflow gauges. That makes the dimensions of input large and the model complex. For reducing the complexity of the ANNs models, clustering neural networks have been proposed to build the rainfall-runoff models.
In this study, Enforced Self-Organizing Map (ESOM), which is based on Self-Organizing Map (SOM), is proposed to improve the performance of streamflow forecasting during flood events. The ESOM increases the mapping space of peak flow in the topological structure of the SOM and improves the accuracy of forecasting peak flow. Then, Learning Vector Quantization (LVQ) is used to adjust the clustering centers of the ESOM further. We have investigated the SOM, ESOM and ESOM+LVQ by using the rainfall-runoff data of the De-Chi Reservoir during flood periods. The results demonstrate the ESOM+LVQ has a good clustering capacity with high efficiency for flood prediction.
|
|---|
