Modeling, forecasting and risk evaluation of wind energy production – FOREWER

On the one hand, state of the art statistical and probabilistic modeling tools (wavelets, stochastic processes) will be applied to the historical weather simulations performed at LMD (consortium partner), in order to understand the multiscale behavior of the wind resource, analyze its variability modes and identify the predictable components of the distribution. On the other hand, powerful statistical learning methods, developed by the statistics group at LPMA (coordinating partner) will be adapted to identify the salient predicting features as well as the connections between renewable power production and the meteorological variables. The statistical forecasting methodology successfully used by LPMA to predict the power consumption curve will be adapted to obtain seasonal and decadal projections of these relationships and produce reliable probabilistic forecasts of the renewable power production taking into account the climate non-stationarity.

During the first 18 month of the project the research has mainly been carried out in four directions.

Medium-term forecasting of the renewable resource. We studied the variability of the wind energy resource in France on monthly to seasonal timescales. We first developed a model for the relationship between the large scale circulation and the surface wind speed. We then applied this model to seasonal forecasts from the European Center for Medium-range Weather Forecasts (ECMWF) in an attempt of forecasting the monthly and seasonal distribution of the surface wind speed.

Short-term forecasting of the renewable power production. Real-time wind power forecast of a wind farm has been considered based on meteorological data. Machine learning technics have been tested and compared to parametric models. The error of prediction of the new method represents less than 1.65% of the installed power.

Integration of the renewable sources into the energy systems and energy markets. We determined the optimal trading policies for a wind energy producer who aims to sell the future production in the open forward, spot, intraday and adjustment markets, and who has access to imperfect dynamically updated forecasts of the future production.

Large-scale statistical modeling of meteorological variables. Meteorological conditions have a direct impact on the French electric power consumption and renewable power production, and being able to characterize the climate at a regional level may help to improve electricity forecasts. We have shown that, regarding the temperature and wind information, the French territory can be described, from a climate point of view, by a few homogeneous regions, which appear to be globally stable for the last 15 years.

The major innovations of the project Forewer are

- Analysis of the medium and long-term probabilistic predictability of the wind resource using state-of-the-art statistical tools.

- The end-to-end approach, which consists in considering the whole chain of wind power production from the modeling and prediction of the renewable resource to the management of the associated risks.

During the first stage of the project, four articles have been written :

B. Alonzo et al., Modelling the variability of the wind energy resource on monthly and seasonal timescales, to be submitted subject to validation by the partners

Z. Tan and P. Tankov, Optimal trading policies for wind power producer, to be submitted subject to validation by the partners.

P. Tankov, Tails of weakly dependent random vectors, Journal of Multivariate Analysis, Vol. 145, 73-86 (2016)

A. Fischer, L. Montuelle, M. Mougeot and D. Picard, Real-time wind power forecast, to be submitted subject to validation by the partners.

In addition, several conference communications on these and other topics related to the project have been made.

For reasons of environment protection and energy security, the share of renewable resources in the global energy supply is now rising at an overwhelming rate. The European Commission has set the target to reach a 20% share of energy from renewable sources by 2020 and further increases of this already ambitious objective will follow.
A large fraction of this growth is to come from wind power. The production of electricity from this resource is both spatially distributed and highly dependent on atmospheric conditions and thus intermittent in nature, leading to challenging planning and risk management problems for the stakeholders of the wind energy industry.

These new challenges, in particular, those related to investment planning and grid integration under the conditions of large-scale wind generation, call for better understanding of the spatial and temporal distribution of the wind resource and wind power production via precise statistical and probabilistic models. Besides, recent advances in climatology show that it may be possible to develop medium and long-term (seasonal to decadal) probabilistic forecasts of the wind power output with a better performance than that of forecasts based on climatological averages, leading to improved risk management tools for wind power producers and grid operators.

The project FOREWER aims to address these crucial issues through a synergy between the statistical and probabilistic methodology and the modern meteorological models. This multidisciplinary public-private partnership brings together mathematicians working on stochastic modeling and risk management, statisticians, and meteorologists from the academic community as well as engineers from the key players of the renewable energy industry.

Our goal is first of all to develop reliable theoretical and numerical models and scenario generators for the wind resource distribution and power output at various spatial and temporal scales with a focus on medium to long term (seasonal to decadal).
We shall then evaluate the potential of these tools for solving the forecasting and risk management problems relevant for the industrial partners of the project, such as the evaluation of the sensitivity of a proposed wind farm to climate variability and optimal placement of wind farms, determination of the required capacity of back-up generators and optimal operation of these assets, and integration of renewable power sources into the grid.

On the one hand, state of the art statistical and probabilistic modeling tools (wavelets, stochastic processes) will be applied to the historical weather simulations performed at LMD (consortium partner), in order to understand the multiscale behavior of the wind resource, analyze its variability modes and identify the predictable components of the distribution. On the other hand, powerful statistical learning methods, developed by the statistics group at LPMA (coordinating partner) will be adapted to identify the salient predicting features as well as the connections between renewable power production and the meteorological variables. The statistical forecasting methodology successfully used by LPMA to predict the power consumption curve will be adapted to obtain seasonal and decadal projections of these relationships and produce reliable probabilistic forecasts of the renewable power production taking into account the climate non-stationarity.

The major innovations of the proposed project are
- Analysis of the medium and long-term probabilistic predictability of the wind resource using state-of-the-art statistical tools.
- The end-to-end approach, which consists in considering the whole chain of wind power production from the modeling and prediction of the renewable resource to the management of the associated risks.
The predictive power of our models will be analyzed in case studies with our industrial partners.