Initial Evaluations of the Performance of an Enhanced Model to Forecast Day-ahead Solar Irradiation

摘要

The growth of photovoltaic power generation in Japan and around the world is generating a strong and continuous demand for better day-ahead forecasts of photovoltaic power and solar irradiation. In this extended abstract we report initial results obtained when investigating potential enhancements of a machine learning based model used to forecast day-ahead solar irradiation. The evaluations were done using one year of hourly data of 8 locations in Kyushu region, and the accuracy of both local and regional forecasts were analyzed. The effect of the proposed enhancements is verified using the original forecast model forecasts as reference. Additionally, since Dec 5th, 2017 the Japan Meteorological Agency, JMA, is publishing its physical model based forecasts of solar irradiation, providing a valuable reference to which compare new forecast methods. Thus, an initial comparison with this new standard is also provided.