Photovoltaic generation systems have been attracting attention in recent years as a countermeasure against the depletion of fossil fuels and global warming. Since PV is affected by the weather and PV power changes rapidly, there is a risk of adversely affect the power grid. To solve this problem, there is a method for predicting solar irradiance using the all-sky image. The method of previous research estimates the solar irradiance from all-sky images, and then predicts the solar irradiance by inputting all-sky images and the estimated solar irradiance into a convolutional neural network (CNN). This study evaluates the impact of the pixel size of all-sky images on estimation accuracy in a CNN-based solar irradiance estimation method.
