Investigating the use of readily accessible climatic data for predicting vertical solar irradiance under sunlit and shaded scenarios

Abstract

For evaluating active and passive energy systems, vertical global irradiance (GVT) data are required. However, unlike horizontal irradiance, GVT measurement is sparingly available, making GVT models an alternative. Also, previous GVT models have been developed using diffuse irradiance (DVT). Nevertheless, estimation methods for determining the DVT are a huge source of computational error in GVT modelling. This study proposed GVT models for two GVT cases (i.e., sunlit and shaded vertical surfaces) based on the vertical direct and reflected irradiance. By separating these cases into two, the error-prone DVT estimation method was omitted. The input variables used were combinations of the ratio of direct normal irradiance to horizontal global irradiance (DNI/GHI), clearness index (Kt) and scattering angle (χ). The method was improved by proposing simple regression and machine learning (ML) models for GVT prediction. ML was also used for feature importance identification. All models were developed using ten minutes of measured data from Hong Kong. Findings show that DNI/GHI and Kt are important variables for modelling GVT for the sunlit and shaded surface, respectively. Also, aside from the linear regression models, the ML models had %RMSE ranging from 8.9 to 13.1% and 12.8 to 20.3% when tested against the 2005 and 2019 to 2020 data, respectively. With most predictions having %RMSE less than 20%, all models (especially the optimised support vector) gave good predictions of GVT. Overall, the proposed models will be useful to building designers and researchers in deriving irradiance data for building energy evaluation.