PROBABILISTIC PREDICTION OF DAM INFLOW WITH UNCERTAINTY IN FORECAST RAINFALL USING BAYESIAN DEEP LEARNING

Abstract

In addition to physical models, statistical models such as deep learning have recently been used to predict high water levels in dams for the purpose of disaster prevention. In this study, we developed a deep learning model to predict the inflow to a dam using the observed and predicted rainfall in the inflow area as input, and examined the appropriate input information through a parametric study. Moreover, a Bayesian Neural Network was proposed as a method to represent the uncertainty of the predicted rainfall and the effect of the model error on the output values, and its effectiveness was verified. Numerical experiments show that the Bayesian Neural Network reflects the uncertainty of the output value according to the amount of training data in the confidence interval of the output, and that the inflow due to extreme rainfall that is not in the training range can be included in the prediction range.