In power systems where large amounts of photovoltaic power generation have been installed, operational planning is based on next-day solar radiation forecasts. In order to realize efficient and stable power system operation, advanced next-day solar radiation forecasting is necessary, and forecasts must not only have average accuracy but also be able to control the risk of rare large outages. In this study, a multiple solar radiation forecast model using ensemble forecast data was constructed by machine learning, and a forecast model that can both improve the average accuracy and suppress the risk of large over-estimation was investigated.
