Poster presentations

Posters at EWEA Wind Power Forecasting 2015

Posters at EWEA Wind Power Forecasting 2015(photo EWEA)

Topic: Users’perspective and experience

Winter forecasting for wind energy: A journey from scientific breakthrough to useful application Emily Wallace
Scientific Consultant,
Met Office, UK

Topic: Numerical weather prediction models

A downscaled ensemble prediction system for offshore weather forecasts Torge Lorenz
PhD Student, Computing,
Uni Research, Norway
Verification and usage of newest generation of seasonal forecast system for wind industry applications Gil Lizcano
R&D Director

Topic: Wind power forecasting models and operational systems

Short-Term Forecasting of Wind Speed and Direction Exploiting Data Non-Stationarity  Alice Malvaldi
PhD Student
University of Strathclye
United Kingdom
Reliable wind speed and wind power generation forecasting using statistical methods  Dr Fiona McGroarty
Lead Researcher NGF Project
Energy Resource Group
A dynamic tool for intraday and extraday forecast windpower  Julien Berthaut-Gerentès
Researcher Engineer
Bidimensional analysis of Wind Energy forecasts including a new Temporal Distortion Index Laura Frías
Researcher, Wind Energy
Short term wind and energy prediction for offshore wind farms using neural networks  Jörg Bendfeld
University of Paderborn
Validation of our own weather forecast system GMS-Profiwind
for 16 wind farms  
Kerstin Schäfer
Technical Manager and Dipl.Informatics
Carsten Albrecht
Creating probabilistic wind farm group forecasts based on uncertainty information for member forecasts using Bayesian model averaging  Hans Georg Beyer
Professor, Engineering
University of Agder

Topic: Grid integration

Wind Power Forecast Accuracy and its Role in Electricity Markets Riccardo Goggi
General Manager
RenEn Italy
Smart Wind Farm Control : Practical use of Probabilistic Wind Power Forecast in a Congested Distribution Grid Gregoire Leroy
R&D Meteorologist
3E, Belgium

Presenter: Emily Wallace, Scientific Consultant, Met Office

Title: Winter forecasting for wind energy: A journey from scientific breakthrough to useful application

Abstract: In 2014 the Met Office revealed remarkable accuracy for seasonal predictions of European winter climate. Probabilistic forecasts of wind, temperature and storminess were now possible ahead of the winter season, bringing with them potential for efficiency and cost savings for the wind energy industry. However there was also a significant mismatch between the exciting research result, and the adoption by end-users.

In this talk I will describe the user-engagement activity and prototype winter 2014/15 forecast service undertaken to address this mismatch. This activity involved wind farm operators, utilities, energy traders and distribution network operators. With each group key questions emerged: What predictions are relevant to the user? How can the user integrate the probabilistic forecasts into their business processes? What actions can they take as a result of the forecast?

Come along to hear what we learned about introducing this exciting science to a new user group, and how the forecasts were actually used.



Presenter: Gil Lizcano, R&D Director, Vortex

Co-authors: Abel Tortosa, Vortex,Melanie Davis, IC3/BSC,Francisco Doblas-Reyes, IC3/BSC


Title: Verification and usage of newest generation of seasonal forecast system for wind industry applications

Abstract: A newest generation of seasonal to decadal (S2D) predictions systems is being developed within the framework of SPECS consortium ( SPECS project initiative joins the international efforts to improve S2D forecast quality and to provide efficient regionalisation tools for reliable, local climate information.

This works presents results and outcomes from SPECS seasonal scales forecast verification work-package for wind industry end-users. The final aim of the analysis is to assess and to delimit wind industry usability of current seasonal forecast products.

A set of wind conditions-calibrated seasonal predictions systems  were verified against gridded reanalysis data, which are employed as proxy observation for wind conditions intrannual variability. The forecast system employed in the analysis were: North America multi-model NMME project, ECMWF System-4 and CanSip system.

The analysis was splitted in different regions with different levels of predictability. Seasonal predictions were analysed in a the frame of a probabilistic approach. Advanced dressing ensemble methods and other statistics technique were employed to optimize information obtained from multi-model data sources.

Results were interpreted in terms of the dialectic  between wind industry critical perspective and climate developers experimental skills improvements . Different cases were used to better visualization of potential applications, forecast quality limitations and evidence of improvement over persistence forecast benchmarks.

The second part of the work, present three  test-cases  of usability of seasonal predictions: regional wind index forecast, offshore O&M applications and  windfarm localization of predictions.


If predictability is the limit of predictions, what is the limit for usage? This question summarize the current status of the development of seasonal forecasting technology applied to different pilot areas, such wind industry.  This works aims to provide quantitative basis to assess current seasonal forecast technology quality and to illustrate usage via some  pilot wind industry usage cases.



Presenter: Torge Lorenz, PhD Student, Uni Research

Co-authors: Idar Barstad, Uni Research Computing

Title: A downscaled ensemble prediction system for offshore weather forecasts

Abstract: A short-term weather forecast for the North Sea is being developed to reduce costs and improve safety of marine operations during installation and maintenance of offshore wind turbines. We apply the mesoscale Weather Research and Forecasting (WRF) Model to downscale the global ensemble prediction system from the European Centre for Medium-Range Weather Forecasts (ECMWF) to a horizontal grid spacing of 3 km in the North Sea. Near-surface wind speed is one of the key variables in determining safety of marine operations. To increase the ensemble spread and improve the probabilistic forecast of near-surface wind speed, we introduce perturbations of the sea-surface temperature (SST) into our downscaled ensemble. The SST perturbations are designed to represent small-scale SST features which are not resolved by the ECMWF model but should be present on the scale of our WRF ensemble. Their amplitude is based on statistical analysis of the Multi-scale Ultra-high Resolution (MUR) Sea Surface Temperature Analysis on different spatial scales. The dynamical downscaling is shown to increase the ensemble spread in near-surface winds. The added value from the SST perturbations is investigated by validating the downscaled ensemble and a control ensemble without additional SST perturbations against satellite measurements of near-surface wind components.

Presenter: Alice Malvaldi, Student, University of Strathclyde

Co-authors: Jethro Dowell, Stephan Weiss, and David Infield

Title: Short-Term Forecasting of Wind Speed and Direction Exploiting Data Non-Stationarity

Abstract: In previous work [1], considering the seasonal cyclo-stationarity has enabled the design of enhanced linear predictors for short term fore- casting of wind speed and direction modelled as a complex valued time series when compared to a stationary assumption. Besides these sea-sonal patterns, wind data also exhibits strong diurnal dependences, which e.g. have previously been exploited successfully in the prediction of solar power [2]. Therefore, this paper extends the analysis in [1, 3] to incor-porate both the seasonal and diurnal variations into the calculation of a multichannel complex valued cyclo-stationary prediction filter that is applied to wind speed and direction data. We demonstrate the benefit of this approach for a large data set from the UK’s Met Office [4].Keywords: Non-stationarity, short-term wind forecast, multichannel pre-dictive filter



Presenter: Dr. Fiona McGroarty, Lead Researcher NGF Project, Dublin Institute of Technology

Co-authors: Thomas Woolmington

Title: Reliable wind speed and wind power generation forecasting using statistical methods

Abstract: Ireland has set an ambitious target of obtaining 40% of electricity consumption from renewable sources by 2020. As a consequence, reliable wind speed and wind power forecasting methods are urgently needed. These methods will need to generate accurate forecasts over a range of time frames from minutes to days. More importantly, the uncertainties in the forecasts must be well defined, as it is this information that will allow wind energy to be a reliable source of power.

Our NGF wind power forecasting method comprises two interlinked components that forecast wind speeds and wind power generation.  A number of different statistical models using (i) machine learning techniques on time series data and (ii) non-gaussian methods using a Levy Index are developed and evaluated for the short (10/30/60/180 minutes) and medium term (24 hours). The models have different advantages depending on the wind profile input and the forecasting horizon, but generally they are accurate to 1 – 2 m/s on average in the short term time frames. The driving factor in the model development has been to better constrain the uncertainties and to be able to state the forecast accuracy for a specified time frame with reliable uncertainties.

These accurate wind speed forecasts are then fed into a wind power generation model that forecasts wind power on the same short and medium terms.

Again, a number of different machine learning models are developed using time series power generation data and the optimum model is selected based on the wind speed inputs and the forecast horizon required.

Having reliable wind power generation forecasts, with well-defined uncertainties will allow the end-users to know their power production in a given time frame with a good deal of confidence. This is important in terms of the end-users O&M and also for trading on the electricity market.

Presenter: Julien Berthaut-Gerentès, Research Engineer, METEODYN

Co-authors: Stephane Sanquer, Jean-Claude Houbart

Title: A dynamic tool for intraday and extraday forecast of windpower

Abstract: Forecast systems are classically classified by their time-horizon: very-short term, short, medium… We focus the interest on two specific groups: the extraday forecast and the intraday forecast. The first one is widely needed whole over the countries, while the second is more country-specific (or even state or region-specific). Another great difference between these systems is the input data: while the extraday power forecast is based on Numercial Weather Forecast, the intraday power forecast uses local online measurements.

Indian market is a bit specific because its regulation allows the providers to update their extraday forecast several times a day. As a result, the limit between short term (extraday) and very-short term (intraday) forecasts is blurred. We present here an operational tool that provides a continuous forecast from +0H to +48H, mixing smoothly the two classical forecast approaches.

The main idea is to complete a NWP+micro CFD downscaling by a persistence model, throughout an Artificial Neural Network (ANN). This provides an optimal mix between the two approaches, the balance between them being horizon-dependent.

Once the system installed and the ANN trained, the forecast tool is 100% automatic: constantly searchable. It automatically uses the last NWP data and the fresher data measured on the wind farm, and provides a complete forecasted time-series from +0H to +48H. Thus, the end-user can use it as an extra-day forecast as well as an intraday forecast system. The performance (nRMSE) naturally exhibits a high dependency on the time-ahead horizon.

Presenter: Laura Frías, Researcher, CENER

Co-authors: Fermín Mallor, Iván Moya

Title: Bidimensional analysis of Wind Energy forecasts including a new Temporal Distortion Index

Abstract: Wind has been the largest contributor to the growth of renewable energy during the early 21st century. However, the natural uncertainty that arises in assessing the wind resource implies the occurrence of wind power forecasting errors which perform a considerable role in the impacts and costs in the wind energy integration and its commercialization. The main goal of this work is to provide a deeper insight in the analysis of timing errors which leads to the proposal of a new methodology for its control and measure. A new methodology is proposed to be considered in the estimation of accuracy as attribute of forecast quality.

This methodology is based on the Dynamic Time Warping principles which obtains the optimal alignment of two time series by applying dynamic optimization to a shortest path problem. In this problem nodes represent possible temporal pairings between the two series, while possible transitions and distances are defined by a recursive function. This function manages which temporal leaps

are allowed and its associated cost to reach a new coupling.

A new dissimilarity measure, the Temporal Distortion Index, among time series is introduced to complement the traditional verification measures found in the literature. Furthermore we provide a bi-criteria perspective to the problem of comparing different forecasts.

Presenter: Jörg Bendfeld, Dipl.-Ing., University of Paderborn

Co-authors: Stefan Balluf, Jorg Bendfeld, Stefan Krauter

Title: Short term wind and energy prediction for offshore wind farms using neural networks

Abstract: Introduction

In response, offshore wind plants become an important contributor of electricity generation from wind energy. The key issue for the stability of the grid is to cope with varying intermittent generation of electrical power delivery. As wind speeds vary with time and location and are different on- and offshore, the level of intermittency for the electricity system depends on the spatial distribution of wind farms. Wind speed is mainly influenced by factors that change over small spatial resolutions such as elevation, roughness and obstacles.

Data from operational offshore wind farms is only available over a short time period. Consequently, it is difficult to extrapolate current data to future deployments.

Predicting the short term wind speed and the resulting energy is of high importance with regards to energy markets or for the wind farm management.

Forecast Methodology

The paper focusses on the wind and energy estimation using feedforward backpropagating neural networks. Neural networks are, simplified, number processing systems, using the input plus bias as well as weights to determine an output. The amount of inputs, as well as outputs, vary from a few to many. But the important part of the network is the relation between the input values, the interconnection.

Introducing the physical processes in the atmosphere responsible for the wind, the key factors are temperature and pressure including their gradients. As the impact of pressure within the atmosphere is one of the main drivers for wind speed it is a main input of the neural network used here.


The knowledge of the estimated energy helps commercial departments trading energy at the energy markets, selling all of the provided energy without any overhead or lack. The paper shows the setup and some results of the forecasting system.

Presenter: Kerstin Schäfer, technical manager and Dipl. Informatics Carsten Albrecht, AL-PRO GmbH

Title: Validation of our own weather forecast system GMS-Profiwind for 16 wind farms

Abstract: In January 2015 AL-PRO started with the Global Microcasting System GMS, a meso scale wind farm yield prediction model. It is calculated on a company owned high performance computing system and is based on the GFS-data. It runs with different resolutions for Europe. The highest resolution is currently made for Germany with a grid spacing of 4km and a forecast horizon of 3 days.

The outstanding benefit of running its own model is the possibility to calculate the yield of a wind farm exactly for the coordinates of the wind farms and to have all model internal parameters available to train the GMS neuronal network, for example wind speed in 150m. Furthermore we are using an immanent post processing System. For the presented study we validated our model for 16 wind farms over 5 month. The results are very good even without correction of the prediction by neural networks.

Additionally, the GMS neural network correction is very efficient. The results of our tests show a significant improvement of the yield results even with a training of just a few months.

In the future the performance of the neural networks can be expected to improve significantly with the availability of more training data, which was not available for the presented study. This will give wind farm operators, utility’s and energy traders the best option to know how much yield they will produce in the next days.

Presenter: Hans Georg Beyer, Professor, University of Agder

Co-authors: Pal Preede Revheim

Title: Creating probabilistic wind farm group forecasts based on uncertainty information for member forecasts using Bayesian model averaging

Abstract: Forecasts of the lumped power output of a group of wind farms spread over a geographic area – and connected to a common grid – are needed for power scheduling and management of the grid. For best applicability, the forecast given should contain information on the probability distribution of the expected power, which requires knowledge on the forecast error characteristics. Best estimation of the distribution of the expected lumped power can be based on the characteristics of the field of single member (individual wind farm) forecasts. A scheme for this – applying Bayesian Model Averaging – should be presented here.

Bayesian model averaging (BMA) is a statistical post-processing method for producing probabilistic forecasts from ensembles, which in this case is the set of single site forecasts. The BMA predictive PDF of the future wind power production of a group of wind farms is a weighted average of single-farm PDFs, where the weights can be interpreted as posterior probabilities and reflect the single site forecasts’ contribution to overall group forecasting skill over a training period.

Such a scheme is applied here to a set of synthetic (modelled power output data based on wind speed measurements and 24h-forecasts) for 43 stations in Norway, separated into 7 groups. The BMA scheme is applied to the different groups. The performance is assessed by inspecting the number of occurrences of actual data outside the confidence limits given by the scheme for different confidence levels.

Presenter: Riccardo Goggi, General Manager, RenEn

Co-authors: Francesca Egidi, Diego De Luca

Title: Wind Power Forecast Accuracy and its Role in Electricity Markets

Abstract: Numerical Weather Prediction (NWP) models provide information on the evolution of wind speed and direction. However, it is commonly known that the evolution of the atmospheric phenomena is conditioned by local factors that modify the velocity profiles forecasted by the NWP.

These phenomena, caused by non-homogeneity of the surface, topographic features, thermal processes and obstacles, are open to misinterpretation from weather providers and create variability that can heavily impact the forecasting quality.

An accurate forecast of wind production by zone is a useful tool to forecast wholesale power prices and it can represent a decisive added value to traders if linked to the knowledge of the other main market variables influencing prices.

Furthermore, an accurate forecast for individual wind farms can improve their production scheduling on the day ahead market and its adjustment on the intra-day markets in order to reduce unbalancing cost. This in turn also minimises network disruption by intermittent renewable sources, thereby benefiting the entire system.

Thanks to our expertise in solar and wind forecasting with over than 6 GW of forecasted power plants in 2015, combined with our hands-on know-how in energy trading, we are probably in a unique position to produce both accurate forecasts and to perform analyses to evaluate the risk connected with the power plant’s market strategies. In our work we take into consideration all relevant variables that can affect local wind forecasts and all elements that can impact short term power prices (e.g., solar power forecasts by zone, electricity demand, thermal power plant availability, available interconnection capacities, etc.).

Through some case studies on the Italian electricity market, we show how the use of accurate wind forecasts and the analysis of market fundamentals, jointly with solar forecasts, are essential for the optimisation of any generation asset, wholesale market participant and TSO.


Presenter: Grégoire Leroy, R&D Meteorologist, iLab, 3E

Co-authors: Rory Donnelly, Patrick Hoebeke, Elvin Lemmens, Babacar Sarr

Title: Smart Wind Farm Control : Practical use of Probabilistic Wind Power Forecast in a Congested Distribution Grid

Distribution system operators (DSOs) today face substantial challenges to integrate large wind power plants on distribution level: since this requires expensive and time-consuming investments to reinforce the network, the development of many of these plants can get postponed or even cancelled. The Belgian DSO Eandis is currently demonstrating how a large wind farm (the Wind aan de stroom project in the port of Antwerp) can be connected in the short term by using intelligent grid management techniques during congestion periods; these methods allow the minimization of curtailment losses. Examples of possible grid management techniques are predictive dynamic line rating as well as active network and demand side management. These techniques are driven by a forecasting service at turbine level integrating probabilistic methods over different forecasting horizons. This service has been demonstrated since spring 2015 as part of the SWiFT (Smart Wind Farm ConTrol) project. The project gathers leading industry (Eandis, 3E, General Electric) and research institutes (iMinds, Ghent University and KU Leuven).
3E is demonstrating an operational probabilistic forecast system covering both very-short (up to 6 hours ahead) and short-term (up to 48 hours ahead) forecast horizons. It is based on combined time series (ARIMA, Kalman Filters) and Numerical Weather Prediction models and provides continuous probabilistic forecasts at turbine level every two hours with a time resolution of 15 minutes.
The presentation will describe the practical use of probabilistic wind power forecasts in this concrete use case and present the early demonstration results.