A Simple Model for Estimating the Diffuse Fraction of Solar Irradiance from Photovoltaic Array Power Output

Abstract

Given the rapid proliferation of data recording equipment for distributed photovoltaic (PV) arrays globally, there exists a new opportunity to use the power output from these systems for the purpose of surface solar radiation assessment. Direct measurements of the beam and diffuse irradiance represent the best methods for producing such assessments, however the equipment required for these observations are expensive and require routine maintenance, which therefore mean that the measurements are quite sparse globally. Satellite derived solar radiation estimates, meanwhile, have global coverage with increasingly fine resolution, but still require surface measurements of radiation in order to assess the performance of their solar radiation estimation algorithms (e.g. Heliostat). Therefore, it is global horizontal irradiance measurements recorded by a pyranometer, which have become the most common measurement of surface radiation. Pyranometers provide accurate surface radiation observations and are relatively inexpensive. As such, models which separate the diffuse and beam components in a global measurement have been discussed and developed vigorously in recent decades, with many modern models now accepted as the state of the art. This paper posits that the power output from PV systems is not altogether different from that recorded by pyranometers, and could be used in place of, or in supplement to, radiation observation equipment. This would greatly increase the density of the surface radiation measurement network, allowing for the many millions of PV systems reporting power output measurements globally to be applied to this purpose. PV system power output has a first order relationship with incoming solar radiation, but is confounded by additional second order interactions such as losses related to temperature, module efficiency, DC-AC conversion, soiling and shading, etc. Recently, research work by the first author has demonstrated that the individual nuances of PV systems can be accommodated through normalisation of their power output to their simulated clear sky performance. This normalised variable is termed the clear sky index for photovoltaics, KPV . We use this value as the primary input to a logistical regression model in place of the traditional input, the clearness index Kt, and explore the use of additional predictor variables to optimise accuracy. PV power output was collected from 18 sites in two Australian cities (Adelaide and Melbourne) in which Bureau of Meteorology solar radiation measurement stations are deployed. This allowed us to fit and test Kt and KPV based models to the observed diffuse radiation, and directly compare these approaches. Surprisingly, initial results suggest a KPV based model has nearly equivalent performance to that of the traditional, pyranometer based Kt model. This paper will explore this relationship more fully, and provide the first simple model available for this purpose