Comparison of different physical models for PV power output prediction

The electricity produced from renewable energy, in particular from wind and photovoltaic plants, has seen exponential rise in the last decade. Consequently, the prediction of power produced from these plants is fundamental for the reliability, safety and stability of the grid. This paper compares three physical models describing the PV cell (corresponding to three-, four- and five-parameter equivalent electric circuit) and two thermal models for the cell temperature estimation (NOCT and Sandia). The models were calibrated and tested towards ten monocrystalline and eight polycrystalline modules installed at SolarTechLab at Politecnico di Milano. The hourly error of the forecasted power output is usually lower than 15 Wh, while NMAE% and WMAE% are in the range of 0.5% and 10%. Low errors, calculated with actual weather conditions, suggest that the implemented models are accurate, but they cannot be directly compared with other approaches which adopt weather forecasts. Results show that there is no clear advantage of using complex models, but the data used for the model calibration mostly affect the model accuracy. It was found that forecasted power output are more accurate using experimental data and Sandia's thermal model in monocrystalline cells type, while for the polycrystalline the data from the manufacturer and NOCT have lower errors.