Advances In Short-Term Wind Power Forecasting With Focus on 'Extreme' Situations - SafeWind
Résumé
Introduction: The integration of wind generation into power systems is affected by uncertainties in the forecasting of the expected power output. Misestimating of meteorological conditions or large forecasting errors (ie phase errors, near cutoff speeds) are costly for infrastructures (ie unexpected loads on turbines) and reduce the value of wind energy for end-users. Existing wind power forecasting approaches focus on the "usual" operating conditions rather than on extreme events. This paper presents the research methodology and first results of the European project SafeWind, which aims at developing extensive research for improving wind modeling and forecasting in challenging or extreme situations. Approach: Due to the variable nature of the wind resource, the large-scale integration of wind power causes several difficulties to the operation and management of a power system. Short term forecasts of wind generation, from a few hours up to a few days ahead, are necessary for the optimal integration of wind generation into power systems. However, existing forecasting approaches focus on the "usual" operating conditions rather than on extreme events. This paper presents an overview of the main results of the 4-year European project SafeWind (FP7) which is coming at its end by 2012. Main body of abstract: The results presented here cover three main axes of Safewind. Firstly, the project aims at improving predictability with focus on extremes at various temporal and spatial scales going from a few minutes to a few days and from the level of wind turbine to the European scale respectively. Although current forecasting technology mainly encompasses deterministic models for wind production, the project develops the concept of complementary tools that can be used jointly to traditional forecasts to assess wind predictability. The project developed: * new forecasting methods for wind generation focusing on uncertainty and challenging situations/extremes (i.e. probabilistic models for ramps timing). * models for "alarming": providing information for the level of predictability in the (very) short-term. They use near-real time online observations for alerts on potential extreme prediction errors and for immediate updates of short-term (0-6h) wind power predictions on regional and local scale; * models for "warning": providing information for the level of predictability in the medium-term (next day(s)). Such tools, based on ensemble weather forecasts and weather pattern identification, can be used to moderate risks in decision making procedures related to market participation, reserves estimation etc. The second axes of the project is to study how new measurement technologies like Lidars can be beneficial for improved evaluation of external conditions, resource assessment and forecasting purposes. Advances to that direction are presented in the paper. At the early stages of wind energy, the focus was on resource assessment, where the aim is to take optimal decisions where to install new wind farms. Nowadays, the revenue of a wind farm may be generated by its direct participation to an electricity market. Prediction errors result to penalties that reduce revenue. The third axes of the project studies how the predictability of a site can be considered as a design parameter when taking decisions for the installation of a new wind farm. It is studied whether a site with lower resource but with higher predictability may be advantageous to select. Conclusion: This paper concludes with a critical discussion on the results. This is the basis for a number of recommendations presented on future R&D directions for improving wind predictability.