Abstract
The paper investigates a total of 725 combinations of decomposition and transposition models for calculating the solar potential of building-integrated photovoltaics. Model outputs were experimentally validated against one-minute data from the six building surfaces – the four facades, the tilted roof, and the southwards pergola – of the Zero Emission Building Laboratory, in Trondheim (Norway). The surrounding built environment was considered through shadow masks and sky view factors. We proposed a new cumulative performance indicator, which was comprehensive of the commonest statistical indicators in validation studies, to identify the models that best fitted the observations, for each surface. Perez2/Temps and Perez1/Perez1 optimally handled the tilted roof and pergola, respectively, but they were outperformed by Perez1/Perez1 and Perez3/Perez1 in case of vertical facades. These model chains were further analysed by mapping the distribution of residuals according to the time and the predictor values. The availability of data points influenced the model chain performance as observed for the fourth trimester (November, December, January) in the monthly analysis and the sunrise/sunset hours in the hourly analysis. Re-parametrizing the models by giving greater weight to these time frames can enhance the model’s reliability. Additionally, larger and/or more dispersed residuals were found for clearness index values lower than 0.40 and for angle of incidence values higher than 50◦ or close to 20◦. Similarly, rooms for improvement for model chain performance were identified in the case of relative airmass values from 5 to 10 and low global horizontal clear-sky irradiance.