ACCURACY IMPROVEMENT OF FLOOD FORECASTING MODEL USING ARTIFICIAL NEURAL NETWORK FOR INEXPERIENCED FLOOD

Abstract

The Artificial Neural Network (ANN) is widely used as a flood forecasting model that can be applied to a wide range of watershed. One of the difficulties of the ANN model is to clarify the prediction accuracy for the inexperienced flood. To achieve good prediction accuracy with the ANN model, setting of network architecture and the learning processes is very important. However, few studies have been conducted to determine appropriate number of hidden units and the number of training calculation. In this study, we used a case study approach to explain how the number of hidden units and training calculation affect to the prediction accuracy, and thus how we can determine them on a reasonable basis. Consequently, we developed the ANN flood prediction model of the Saba-River, Yamaguchi prefecture, on the proposed approach, and confirmed good accuracy between observed and predicted flood water level at the recent biggest flood events.