Establish Neural Network Real-time Daily-flow Discharge Forcasting Model

• Main Authors: Ching-Jin Wu, 吳青俊
• Other Authors: none
• Format: Others
• Language: zh-TW
• Published: 2002
• Online Access: http://ndltd.ncl.edu.tw/handle/ejp687

碩士 === 逢甲大學 === 土木及水利工程所 === 90 === With dramatic variations both in temporal and spatial scale, we keep water resource to build a reservoir in Taiwan. The reservoir that prevents floods or stores water resources depends on accuracy of stream flow forecasting. But discharge is temporal variation, how to know discharge’s identities and then predict it become as a very important topic. In this research, we try to use Artificial Neural Network to establish Recurrent Neural Network (RNN) that processes real-time stream flow forecasting. It can forecast stream flow in a basin by keeping learning and correcting. In tradition, the Back-Propagation Network (BPN) forecasts stream flow with good precision as usually, but it has to retrain a new model when there is new information about stream flow. It wastes a lot of time and causes inconvenience in use. Due to this reason, we try to use Online Learning method to build RNN model that renew the weights and show the model’s change more confidently and fast when there is new information about stream flow. In this research, we use RNN that provides a real-time prediction method to control water resources process daily stream flow forecasting of Wu-Xi basin, and use the daily stream flow data of Da-Do bridge stage in the downriver to represent the hydrologic situation of Wu-Xi basin. The results of precision in stream flow forecasting, shortening time in building model to predict it, or fast convergence are better by using RNN than BPN. It is a useful and powerful method in real-time prediction.