レーダデータを用いたニューラルネットワークによる発電用ダム上流域の地上雨量分布推定

Abstract

This paper describes an application of a neural network method for estimating the ground rainfall distribution from radar echo data. A neural network system for this purpose is developed through a case study on a dam for hydro-power plant located the upper district of the Hida River in Central Japan. We use the neural network comprised of three layers; an input layer, a hidden layer and an output layer. The input data to the neural network are a radar echo amount observed in each radar mesh and x-y coordinates showing its location. The output from the neural network is the estimated ground rainfall amount. Thus, the neural network has three nodes for the input layer and a single node for the output layer. A set of three nodes is adopted for the hidden layer. It is found that the estimating system yields goods results for the ground rainfall distribution. Further, the height of the ground corresponded with to the radar mesh is taken as an additional input datum. It is also found from our investigations that the estimating accuracy of the neural network is improved by the additional datum.