Search documents

Browse topics

Document details

Could a Physical Model Be Used to Forecast PV?
O.T. Huxley, C. Allen, J. Taylor, J. Briggs, A.M. Everard, A. Buckley
Photovoltaic (PV), Forecasting, Modelling / Modeling
PV System Performance and Integration
Subtopic: Solar Resource and Forecasting
Event: 33rd European Photovoltaic Solar Energy Conference and Exhibition
Session: 6BV.3.19
2492 - 2497
ISBN: 3-936338-47-7
Paper DOI: 10.4229/EUPVSEC20172017-6BV.3.19
0,00 EUR
Document(s): paper


We investigate the use of a deterministic approach to modelling PV using a physical model and assess whether this approach is suitable for forecasting PV. We find that on an individual system basis for hourly data, the physical model is not appropriate for modelling PV because the errors inherent in a physical model are too large at this scale. However, on a national or regional scale, or for daily or weekly data, the physical model offers a fast,easy and viable method for operational forecasting of PV generation. The physical model achieved an 2 of 0.77, a mean normalised RMSE of 36%, and a mean bias error of -0.002 kWh. When the data was aggregated spatially and temporally, the performance of the model was found to improve significantly with the scale of aggregation. The performance of the model under different sky conditions was analysed and the model was found to perform best under very cloudy or very clear conditions, although the model typically overestimates the PV generation for clearer conditions. A commercial irradiation forecast was analysed and compared with ground based observations, leading to an 2 of 0.88 and a mean normalised RMSE of 34%. This error is comparable with the error in the PV model and the two errors combined are deemed to be use-prohibitive for the applications considered.