EVALUATION OF THE INTERANNUAL ACCURACY OF DAY-AHEAD FORECASTS OF SOLAR IRRADIATION FOR KYUSHU IN JAPAN

Abstract

Kyushu region in Japan is an area of interest in power systems analyses, with low power demand, and high levels of photovoltaic power penetration. We present the interannual variability of day-ahead forecasts of solar irradiation for Kyushu, using four year of data. Forecasts were done with a method based on machine learning. The results show that the maximum variation of the root mean square error reached 10.5%. Monthly, the mean forecast errors of February, May and September had little interannual variability. The results indicate that there are months in which the accuracy can be estimated with good confidence.