Bias-Correction of Diffuse Solar Irradiance Modeled Through Parametric Models

Abstract

In this paper, a bias-correction of clear-sky models in prediction of the diffuse solar irradiance is presented. High quality ground date from two networks is used for model input and validation: aerosol optical depth from AERONET and diffuse solar irradiance from BSRN. Four established clear-sky models are tested: REST2, Yang, PS and MAC2. Two types of bias-corrections are developed: i) a correction for multiple aerosol scattering (MSC), accounting for extinction at high turbidity, and ii) a general regression on the model errors as a function of Sun elevation angle (h) and turbidity (β) correction (MOS). While the impact of MSC proves marginal, the MOS correction shows good results. Two versions of MOS are developed, one on the global dataset and a site-specific adaptation. Both versions bring significant improvements over the original implementation of the models, while the site-specific version reduces the errors with a few additional percentage points. The aggregate nRMSE of the REST2 model is reduced from 17.8% (no correction) to 11.3% (global MOS) and 9.4% (site-specific MOS).

Short-Term PV Power Forecasting Based on Sky Imagery. A Case Study at the West University of Timisoara

“…[25] argued, many parametric clear-sky solar irradiance models behave qualitatively similarly to SIM.v2 and REST2 in what regards the sensitivity to changes in  . The different behavior of the PS model might be the source of its surprising performance in some cases [25].…”