International audienceForecasting of the solar irradiance is a key feature in order to increase the penetration rate of solar energy into the energy grids. Indeed, the anticipation of the fluctuations of the solar renewables allows a better management of the production means of electricity and a better operation of the grid-connected storage systems. If numerous methods for forecasting the mean of the solar irradiance were recently developed, there are only few works dedicated to the evaluation of prediction intervals associated to these point forecasts. Time series of solar irradiance and more specifically of clear sky index show some similarities with that of financial time series. The aim of this paper is to assess the performances of a commonly used combination of two linear models (ARMA and GARCH) in econometrics in order to provide probabilistic forecasts of solar irradiance. In addition, a recursive estimation of the parameters of the models has been set up in order to provide a framework that can be applied easily in an operational context. A comprehensive testing procedure has been used to assess both point forecasts and probabilistic forecasts. Using only the past records of the solar irradiance, the proposed model is able to perform point forecasts as accurately as other methods based on machine learning techniques. Moreover, the recursive ARMA-GARCH model is easier to setup and it gives additional information about the uncertainty of the forecasts. Even if some strong assumption has been made regarding the statistical distribution of the error, the reliability of the probabilistic forecasts stands in the same order of magnitude as other works done in the field of solar forecasting

No abstract

Abstract-Forecasts of wind power generation in their probabilistic form are a necessary input to decision-making problems for reliable and economic power systems operations in a smart grid context. Thanks to the wealth of spatially distributed data, also of high temporal resolution, such forecasts may be optimized by accounting for spatio-temporal effects that are so far merely considered. The way these effects may be included in relevant models is described for the case of both parametric and nonparametric approaches to generating probabilistic forecasts. The resulting predictions are evaluated on the real-world test case of a large offshore wind farm in Denmark (Nysted, 165 MW), where a portfolio of 19 other wind farms is seen as a set of geographically distributed sensors, for lead times between 15 minutes and 8 hours. Forecast improvements are shown to mainly come from the spatio-temporal correction of the first order moments of predictive densities. The best performing approach, based on adaptive quantile regression, using spatially corrected point forecasts as input, consistently outperforms the state-of-theart benchmark based on local information only, by 1.5%-4.6%, depending upon the lead time.Index Terms-renewable energy, prediction, decision-making, power systems operations, offshore.

Accurate wind power forecasts highly contribute to the integration of wind power into power systems. The focus of the present study is on large-scale offshore wind farms and the complexity of generating accurate probabilistic forecasts of wind power fluctuations at time-scales of a few minutes. Such complexity is addressed from three perspectives: (i) the modeling of a nonlinear and non-stationary stochastic process; (ii) the practical implementation of the model we proposed; (iii) the gap between working on synthetic data and real world observations. At time-scales of a few minutes, offshore fluctuations are characterized by highly volatile dynamics which are difficult to capture and predict. Due to the lack of adequate on-site meteorological observations to relate these dynamics to meteorological phenomena, we propose a general model formulation based on a statistical approach and historical wind power measurements only. We introduce an advanced Markov Chain Monte Carlo (MCMC) estimation method to account for the different features observed in an empirical time series of wind power: autocorrelation, heteroscedasticity and regime-switching. The model we propose is an extension of Markov-Switching Autoregressive (MSAR) models with Generalized AutoRegressive Conditional Heteroscedastic (GARCH) errors in each regime to cope with the heteroscedasticity. Then, we analyze the predictive power of our model on a one-step ahead exercise of time series sampled over 10 min intervals. Its performances are compared to state-of-the-art models and highlight the interest of including a GARCH specification for density forecasts.

Offshore wind fluctuations are such that dedicated prediction and control systems are needed for optimizing the management of wind farms in real‐time. In this paper, we present a pioneer experiment – Radar@Sea – in which weather radars are used for monitoring the weather at the Horns Rev offshore wind farm, in the North Sea. First, they enable the collection of meteorological observations at high spatio‐temporal resolutions for enhancing the understanding of meteorological phenomena that drive wind fluctuations. And second, with the extended visibility they offer, they can provide relevant inputs to prediction systems for anticipating changes in the wind fluctuation dynamics, generating improved wind power forecasts and developing specific control strategies. However, integrating weather radar observations into automated decision support systems is not a plug‐and‐play task, and it is important to develop a multi‐disciplinary approach linking meteorology and statistics. Here, (i) we describe the settings of the Radar@Sea experiment, (ii) we report the experience gained with these new remote sensing tools, (iii) we illustrate their capabilities with some concrete meteorological events observed at Horns Rev and (iv) we discuss the future perspectives for weather radars in wind energy. Copyright © 2013 John Wiley & Sons, Ltd.