As global awareness of climate change increases, energy transition policies toward carbon neutrality are accelerating. Solar power generation is heavily influenced by weather and seasons. There are not many studies that reflect the circadian cycle and the year in solar energy prediction. In this study, major weather and time variables are analyzed using Pearson correlation coefficient and Spearman correlation coefficient. Meteorological variables used include solar radiation, sunlight time, temperature, cloud cover, precipitation, ground temperature, wind speed, humidity, and time. Based on correlation analysis, beneficial features are selectively designed and used for model learning. Time periodicity was encoded using sine-cosine time embedding. As a result of the experiment, applying the Spearman correlation coefficient showed lower prediction error and higher explanatory power compared to Pearson-based models in both single deep learning model and ensemble configuration. Among them, the stacking ensemble model showed the best performance.
Ha et al. (Sat,) studied this question.