Impact of Daily Irradiance Profiles on Intra-Day Solar Forecasting

The analysis of solar energy integration requires capturing the temporal variability of solar irradiance, which can highly increase the errors of solar forecasting models. In this paper, a classification of day types for solar energy applications is investigated and the impact on intra-day solar generation forecasting is assessed. The proposed classification approach uses unsupervised learning based on a combination of self-organized maps and mean-shift clustering with six location-independent metrics related to irradiance variability and energy yield. Two types of forecasts (a deterministic forecast and a probabilistic one) are used as a basis to illustrate the impact of the resulting daily irradiance profiles. The forecasts are emulated by adding white noise to historic hourly irradiance observations and their performance is compared to state-of-the-art benchmarking models. The results illustrate the magnitude of dispersion between intra-hour observations and hourly forecasts, where deterministic forecasts observed 16% to 64% of deviation and probabilistic forecasts observed prediction intervals from 11% to 36%, depending on the daily typology. The findings of this study provide useful information on the impact of intra-day irradiance variability on the performance of forecasting models and how these can be adapted based on each day type to support the integration of solar generation in electricity systems.