Long-Term Solar Irradiance Forecasting

The past decade has been characterized by considerable increase of the penetration level of solar photovoltaic systems in energy systems throughout the world. At the same time, solar irradiance has an intermittent nature. Thus, the efficient management of existing and new solar photovoltaic systems requires an accurate forecasting system of solar irradiance. The purpose of the paper is to develop and validate a long-term forecasting model for solar irradiance. This purpose is achieved by applying of clustering method and standard mathematical statistics. The modeling includes pre-processing of historical data used for forecasting and post-processing of the types of days resulted from the clustering analysis. Historical data include solar irradiance and sky coverage by clouds. Pre-processing supposes bi-normalization of the solar irradiance in time and amplitude, as well as clustering, and post-processing supposes denormalization to get the actual values. Error metrics and confusion matrix indices have been used to assess the accuracy of the proposed forecasting method. Four different model variants have been considered, which differ by pre-processing approach of initial data. The comparison of these model variants shows that for better accuracy it is required to use seasonality aspects of solar irradiance. The main result of paper is the created model, which can be used for the solar irradiance forecast with acceptable accuracy for this type of forecasting and for generating of the types of days for different annual scenarios. The importance of paper results consists in the possibility of using of these scenarios for feasibility assessment of the solar photovoltaic system and identifying of the best solutions for their integration in the energy system.