Photovoltaic and Solar Forecasting: State of the Art
IEA‐PVPS T14‐01: 2013. International Energy Agency Photovoltaic Power Systems Programme, (October 2013)Abstract
The field of solar and photovoltaic (PV) forecasting is rapidly evolving. The current report provides a
snapshot of the state of the art of this dynamic research area, focusing on solar and PV forecasts for
time horizons ranging from a few minutes ahead to several days ahead. Diverse resources are used to
generate solar and PV forecasts, ranging from measured weather and PV system data to satellite and
sky imagery observations of clouds, to numerical weather prediction (NWP) models which form the
basis of modern weather forecasting. The usefulness of these resources varies depending on the
forecast horizon considered: very short‐term forecasts (0 to 6 hours ahead) perform best when they
make use of measured data, while numerical weather prediction models become essential for
forecast horizons beyond approximately six hours. The best approaches make use of both data and
NWP models. Examples of this strategy include the use of NWP model outputs in stochastic learning
models, or the use of measured data for post‐processing NWP models to correct systematic
deviations between NWP model outputs and measured data.
Benchmarking efforts have been conducted to compare the accuracy of various solar and PV forecast
models against common datasets. Such benchmarking is critical to assessing forecast accuracy, since
this accuracy depends on numerous factors, such as local climate, forecast horizon and whether
forecasts apply to a single point or cover a wide geographic area. In the latter case, which is often the
main interest of electric system operators, higher accuracies can be achieved since random errors at
distant locations tend to be largely uncorrelated and to partially cancel out.