A Study of Models Combination for Global Clear Sky Irradiance Models

Clear sky global horizontal irradiance (GHIcs) defines the theoretical maximum irradiance reaching a horizontal surface and are often derived from semi-empirical or physical based clear-sky models. In this study we demonstrate a naive model combination method of global clear sky models for improved estimation of GHIcs. To be specific, 10 best performing beam and diffuse clear sky models are each selected from pervious study [1] and are paired to create 97 combinations of global clear-sky models. By including 18 standalone global clear sky models (distinguish from individual beam and diffuse combined models), this study compares 115 global clear sky models under 100 worldwide solar measurement ground stations. After rigorous data quality control and clear sky detection, 18.7 million 1-min time data points (between 2015-01-01 and 2019-09-30) are used to evaluate all 115 models. Principal Component Analysis (PCA) ranking procedure is then employed to aggregate 12 error metrics and provides the overall ranking scores. The top 3 world-wide performing global clear-sky models are MMAC-v1_IQBAL-C, MMAC-v2_PSIREST and REST2-v5, i.e., not the combination of the best beam and diffuse models. Such results may be due to the over- or under- estimation of different beam and diffuse models. Another interesting finding is the REST2-v5 model which was also listed in the top 10 global clear sky models in previous study. In all, most combined models achieve greater performance (higher PCA ranking scores) than the original ones.