This paper proposes a fluctuation determination method based on CNN. Previous research has reported that prediction accuracy for solar irradiance can be improved by learning solar irradiance prediction models using data with fluctuations and data without fluctuations, respectively and then switching the solar irradiance prediction model according to the fluctuation determination model. However, the fluctuation determination model is in an ideal state. This paper first estimates cloud coverage from all-sky images. Then, the fluctuation determination model based on CNN that uses all-sky images and cloud coverage as input is proposed. As a result of fluctuation determination model inputting cloud coverage, true negative rate was low as 10.1% although accuracy was high as 92.4%. The prediction accuracy for 15 minutes ahead solar irradiance was improved 5.4% compared to the conventional method.
