Cloud Distribution Forecasting Model Using Ground Altitude Information and CNN

This paper describes a cloud distribution forecasting model based on deep convolutional neural network (CNN) for application of forecasting photovoltaic power generation (PV) output. PV output information is used for stable operation of the electricity grid and electricity trading markets. It is important to forecast cloud distribution because PV output varies mainly affected by clouds position and irradiation. In our previous research, satellite images were used for the input of the forecasting model. However, clouds appear and disappear from time to time by some factors such as topography, temperature, and humidity. In this paper, ground altitude information which is geopotential altitude is added to the input of the forecasting model to improve the forecasting accuracy. Geopotential altitude is a height referenced to earth’s mean sea level. Because the altitude is provided as two-dimensional data, it is suitable for the forecasting model which treats with images. The forecasting results with August dataset show some improvements in terms of the forecasting accuracy.