Wind power forecasting (more)


  • Point forecasting of wind power (for horizons from few minutes to several days ahead):
  • I have there considered a number of models based on physical, statistical concepts, and the combination of both. Basically, physical models permit to predict relevant meteorological parameters that influence wind power generation, while statistical dynamical models are then employed for modeling the conversion of meteorological parameters into power, as well as for correcting/improving on the local specific dynamics. Combination of various forecasts also has my interest, since combination often permit to obtain more optimal forecasts in a least square error sense.

  • Probabilistic forecasting of wind power (for the same horizons):
  • Point forecasts only inform about the conditional expectation of wind power generation for each look-ahead time, given the chosen model, information set available at current time, and the estimated model parameters. Even if improving our models, while having all potential data for estimating parameters, there may always be some intrinsic uncertainty in the wind power generation process. One should then accept that only a fully probabilitisc view of the forecasting problem will permit to make optimal decisions, both in terms of power system management of in terms of electricity trading. Consequently in my research, I concentrate on developing and applying methods appropriate for the probabilistic forecasting of wind power generation. I also contribute to showing the additional value of probabilistic vs. point forecasts, through my own work or through collaborations.

  • Scenario forecasts (or trajectories) of wind power:
  • The methodologies that have been developed until recently for the probabilistic forecasting of wind power generation focus on providing information about forecast uncertainty, but for each look-ahead time only, independently of the other look ahead times. Similarly, these methods almost always overlook the spatial dependencies in power generation over a region. Wind energy generation being based on complex meteorological processes, wind power fields obviously are correlated in space and in time. Such information about space-time dependencies may be crucial for a number of decision-making processes that rely on wind power forecasts, like congestion management or optimal power system scheduling for instance. I have therefore been looking at methods for generating scenario forecasts which permit to inform about the uncertainty of wind power forecasts for various locations and look-ahead times, and also about its space-time dependencies. Such forecasts take the form of scenarios of wind power generation (also called space-time trajectories), either based on ensemble forecasts of meteorological variables, or using copula-based statistical methods.

  • Communication of wind power forecasts:
  • Researchers always enjoy working on more and more complex methods and products. It is crucial however that the forecasts that are generated by these methods are used (and used appropriately) by the various forecast users. Probabilistic forecasts and scenario forecasts may be very subtle to appraise and optimally integrate in decision-making processes, even if being open-minded to the world of probability and statistics. A number of researchers have shown that the way forecasts are communicated may influence the manner they are subsequently used. I am therefore interested in this question and try to look at optimal ways of communicating wind power forceasts. More than thinking about colours and forecasts format, this may concern the idea of communicating risk indices, event-based forecast probabilities, utility-based optimal point forecasts, etc.