INPUT DATA SELECTION FOR DAM INFLOW PREDICTION IN SNOW DOMINANT REGION USING ARTIFICIAL NEURAL NETWORK

Abstract

 This paper describes a detailed modeling process of Artificial Neural Network to estimate dam inflow of snow dominant region. During the modeling procedure, the sensitivity of training parameters and selection of the most necessary input data set were carefully surveyed. It was proved that smaller batch size and larger number of epoch during a training process provide improved training results as well as testing results. In the case of hydrological modeling for Naramata Dam Basin, 24 input data were selected as the most efficient input data set, which are precipitation, snow depth, wind speed, and temperature of the past 6-days. However, the sensitive of model performance related to the input data number was not significant, once the most efficient input data are included.