Flood discharge forecasting with Neural-Network Model

• Main Authors: Chun-Yai Huang, 黃群岳
• Other Authors: An-Pei Wang
• Format: Others
• Language: zh-TW
• Published: 2000
• Online Access: http://ndltd.ncl.edu.tw/handle/47548241556140959332

碩士 === 中原大學 === 土木工程學系 === 88 === Neural network model (NNM) was developed to analyze and forecast the flood discharge of the Shihmen basin in Taiwan during the typhoon periods. The NNM consists of four layers with one input layer, two hidden layers, and one output layer. Data of rain stations at catchment of shihmen was taken as input data, and inflow discharge of shihmen reservoir as output data. Data of rain and discharge of the first 4 typhoons was taken as training patterns of NNM; other data was used to verify the flood discharge and time of flood peak of shihmen basin. The weight of NNM was modified by Back Propagation Network (BPN) to build a model of the rain-runoff system. Then time of flood peak and quantity of discharge forecasted by this model can be taken as the operation rules of shihmen reservoir. The NNM compared with the ARX model. It is discovered that ARX model with na= 4, nb= [2 4 3 3 3 3 4 4 4 2], nk= [0 1 2 3 3 4 6 5 6 4], has the excellent results. In forecasting 1,2,3 hours flood discharge, both methods having good results.