We have demonstrated a statistically rigorous method of constructing a closed form function model for the diffuse fraction. Additionally, we have shown how it, along with an innovative quadratic programming formulation, can be used to identify values that have a high probability of being infeasible. The model has been deemed suitable for modelling in general since we checked it for a number of locations in different climates. We have checked the data cleaning capability for other locations and it performed well, but we have only used Geelong as an example. We continue to work on improvements to this modelling in the following ways:
• it is not certain if we should be using the 5% limit for probabilities to identify the outliers for all locations,
• we need to confirm the use of the generic model for more locations, and also refine its construction.
• we have in Boland et al. (2001) identified other predictors to enhance the fit, including solar altitude and daily clearness index. The next section deals with a preliminary discussion about the identification procedure that we are presently undertaking.