Modeling global vertical solar illuminance on building façades using adaptive neuro-fuzzy inference systems under all-sky conditions

Abstract

Global vertical illuminance on building façades is used in daylighting analysis, façade performance assessment, and energy-efficient building design. However, direct measurements are rarely available at conventional meteorological stations, limiting assessments involving different orientations, sky conditions, and interannual variability. This study develops an ANFIS-based methodology to estimate global vertical illuminance for the four cardinal orientations (north, east, south, and west) using only commonly available radiometric and geometric variables: direct normal irradiance (DNI), diffuse horizontal irradiance (DHI), and the cosine of the incidence angle on each façade. Models were trained using 115,250 records from 2024 and evaluated using 750,294 records across six independent years (2019–2023 and 2025) to assess temporal robustness. Results show that the models adequately reproduce global vertical illuminance for all façades under all-sky conditions. Over the six independent years, normalized root mean square error (nRMSE) ranged from 9.91% to 22.60%, and normalized mean absolute error (nMAE) ranged from 6.44% to 15.83%. The largest errors occurred for the north-facing façade, whereas the east-, south-, and west-facing façades maintained nRMSE values of 16.50% or lower. The methodological contribution lies in directly estimating LGV from commonly available radiometric variables and façade incidence geometry, without requiring vertical irradiance measurements, explicit sky classification, or intermediate luminous-efficacy calculations. These results support the use of the proposed approach to estimate LGV at the study site when direct vertical measurements are unavailable. Transferability to other locations should be assessed through independent validation.