Wind power generation is a nonlinear and bounded process. In addition, the fact that wind generation is directly related to wind speed, the characteristics of this process are constantly – though slowly – changing with time: it may be seen as nonstationary. All these aspects make the forecasting of wind power generation a complex task, whatever the time-scale considered (ie. from few minutes to few days ahead). Today, wind power forecasting is recognized as a research field of its own, with connections to meteorology, mathematical modeling, and power system engineering. Forecasts of wind generation are used on a daily basis by wind power producers, energy traders, maintenance planners, regulating authorities, etc. As wind generation takes a more and more significant part of the electricity mix, using forecasts will become a basis to decision-making
A specificity of wind power generation is its highly volatile and hardly predictable nature. In addition, many researchers have shown that optimal decisions in terms of trading or management of wind power cannot be made from a single forecast only. This is the reason why a large part of research efforts are focused on probabilistic forecasting (see Figure below), spatio-temporal aspects of forecast uncertainty, and stochastic decision-making.
At DTU Compute, DynSys, the research related to the wind power application deals with these various topics:
-
Point forecasting of wind power (for horizons from few minutes to several days ahead)
-
Spatio-temporal modeling of wind generation at the level of a region or a country (as well as the forecast uncertainty)
-
Probabilistic forecasting of wind power (up to few days ahead): quantile forecasting, prediction intervals and density forecasts, or alternatively scenarios
-
Skill forecasting, indicating the confidence to have in provided forecasts