Real-Time Flood Runoff Prediction by Fuzzified Neural Network

Abstract

A neural network is developed to forecast flood discharge. Network training is conducted using back-propagation with simulated annealing, where rainfall and discharge data as input and discharge data as output. These hydrologic observation data are generally included in various types of errors, so we applied fuzzy theory to these data including errors. After training is completed, the neural network is used to forecast for 3 and 6 hours ahead using only the current observation data as input. The authors have shown that fuzzified neural network is capable of learning the complex flood runoff process and can get reasonable estimations.