• Energy Research

The relationship between the wind speed derived from the outputs of a numerical-weather-prediction model and from observations is explored using statistical and machine-learning models. Eight years of wind-speed measurements at a height of 10 m (from 2010 to 2017) from 171 stations spread over mainland France and Corsica are used for reference. Operational analyses from the European Center for Medium Range Weather Forecasts (ECMWF) provide the model information not only on the surface flow, but on other aspects of the atmospheric state at the location (or above) each station. In a first step, a large number of explanatory variables are used as input to several models (linear regressions, k-nearest neighbours, random forests, and gradient boosting). The modelled wind speed in the ECMWF analyses, by itself, has root-mean-square errors over all stations distributed widely around a median of 1.42 m s $$^{-1}$$ . Using statistical post-processing and making use of a historical dataset for training, the median of the root-mean-square errors at all stations can be reduced down to 1.07 m s $$^{-1}$$ when modelled with linear regressions, and down to 0.94 m s $$^{-1}$$ with the machine-learning models (random forests or gradient boosting). Yet more significant decreases are found for coastal stations where the errors are largest. The random-forest models are further explored to reduce the list of explanatory variables: a list of 25 explanatory variables, mainly consisting of flow variables (wind speed, velocity components, horizontal gradients of geopotential on different isobaric surfaces, wind shear between 10 and 100 m) and including marginally some temperature variables, appears as a good compromise between performance and simplicity. Finally, as a preliminary test for further work, the relation thus captured between the model outputs and the observed wind speed at a given time is applied to forecasts of the numerical-weather-prediction model, for lead times up to 24 h. The machine-learning model is found to be essentially as relevant on the forecasts as it was on the analyses, encouraging further use and development of these approaches for local wind-speed forecasts.

We build and evaluate a probabilistic model designed for forecasting the distribution of the daily mean wind speed at the seasonal timescale in France. On such long-term timescales, the variability of the surface wind speed is strongly influenced by the atmosphere large-scale situation. Our aim is to predict the daily mean wind speed distribution at a specific location using the information on the atmosphere large-scale situation, summarized by an index. To this end, we estimate, over 20 years of daily data, the conditional probability density function of the wind speed given the index. We next use the ECMWF seasonal forecast ensemble to predict the atmosphere large-scale situation and the index at the seasonal timescale. We show that the model is sharper than the climatology at the monthly horizon, even if it displays a strong loss of precision after 15 days. Using a statistical postprocessing method to recalibrate the ensemble forecast leads to further improvement of our probabilistic forecast, which then remains sharper than the climatology at the seasonal horizon.

Abstract The need to have access to accurate short term forecasts is essential in order to anticipate the energy production from intermittent renewable sources, notably wind energy. For hourly and sub-hourly forecasts, benchmarks are based on statistical approaches such as time series based methods or neural networks, which are always tested against persistence. Here we discuss the performances of downscaling approaches using information from Numerical Weather Prediction (NWP) models, rarely used at those time scales, and compare them with the statistical approaches for the wind speed forecasting at hub height. The aim is to determine the added value of Model Output Statistics for sub-hourly forecasts of wind speed, compared to the classical time series based methods. Two downscaling approaches are tested: one using explanatory variables from NWP model outputs only and another which additionally includes local wind speed measurements. Results of both approaches and of the classical time series based methods, tested against persistence on a specific wind farm, are considered. For both hourly and sub-hourly forecasts, adding explanatory variables derived from observations in the downscaling models gives higher improvements over persistence than the benchmark methods and than the downscaling models using only the NWP model outputs.

Transmission system operator (TSOs) need to project the system state at the seasonal scale to evaluate the risk of supply-demand imbalance for the season to come. Seasonal planning of the electricity system is currently mainly adressed using climatological approach to handle variability of consumption and production. Our study addresses the need for quantitative measures of the risk of supply-demand imbalance, exploring the use of sub-seasonal to seasonal forecasts which have hitherto not been exploited for this purpose. In this study, the risk of supply-demand imbalance is defined using exclusively the wind energy production and the consumption peak at 7 pm. To forecast the risks of supply-demand imbalance at monthly to seasonal time horizons, a statistical model is developed to reconstruct the joint probability of consumption and production. It is based on a the conditional probability of production and consumption with respect to indexes obtained from a linear regression of principal components of large-scale atmospheric predictors. By integrating the joint probability of consumption and production over different areas, we define two kind of risk measures: one quantifies the probablity of deviating from the climatological means, while the other, which is the value at risk at 95% confidence level (VaR95) of the difference between consumption and production, quantifies extreme risks of imbalance. In the first case, the reconstructed risk accurately reproduces the actual risk with over 0.80 correlation in time, and a hit rate around 70–80%. In the second case, we find a mean absolute error (MAE) between the reconstructed and real extreme risk of 2.5 to 2.8 GW, a coefficient of variation of the root mean square error (CV-RMSE) of 3.8% to 4.2% of the mean actual VaR95 and a correlation of 0.69 and 0.66 for winter and fall, respectively. By applying our model to ensemble forecasts performed with a numerical weather prediction model, we show that forecasted risk measures up to 1 month horizon can outperform the climatology often used as the reference forecast (time correlation with actual risk ranging between 0.54 and 0.82). At seasonal time horizon (3 months), our forecasts seem to tend to the climatology.

An avenue for modelling part of the long-term variability of the wind energy resource from knowledge of the large-scale state of the atmosphere is investigated. The timescales considered are monthly to seasonal, and the focus is on France and its vicinity. On such timescales, one may obtain information on likely surface winds from the large-scale state of the atmosphere, determining for instance the most likely paths for storms impinging on Europe. In a first part, we reconstruct surface wind distributions on monthly and seasonal timescales from the knowledge of the large-scale state of the atmosphere , which is summarized using a principal components analysis. We then apply a multi-polynomial regression to model surface wind speed distributions in the parametric context of the Weibull distribution. Several methods are tested for the reconstruction of the parameters of the Weibull distribution , and some of them show good performance. This proves that there is a significant potential for information in the relation between the synoptic circulation and the surface wind speed. In the second part of the paper, the knowledge obtained on the relationship between the large-scale situation of the atmosphere and surface wind speeds is used in an attempt to forecast wind speeds distributions on a monthly horizon. The forecast results are promising but they also indicate that the Numerical Weather Prediction seasonal forecasts on which they are based, are not yet mature enough to provide reliable information for timescales exceeding one month.