Back to the programme printer.gif Print

Delegates are invited to meet and discuss with the poster presenters in this topic directly after the session 'Forecasting: Maximizing grid deliverability and leading your business processes to profitability' taking place on Thursday, 13 March 2014 at 11:15-12:45. The meet-the-authors will take place in the poster area.

Lazaros Exizidis University of Mons, Belgium
Lazaros Exizidis (1) F P Zacharie De Greve (1) Francois Vallee (1) Jacques Lobry (1)
(1) University of Mons, Mons, Belgium

Printer friendly version: printer.gif Print

Presenter's biography

Biographies are supplied directly by presenters at EWEA 2014 and are published here unedited

Lazaros Exizidis has studied Electrical and Computer Engineering at the Aristotle University of Thessaloniki. His specialization was in power engineering while his master thesis, "Dynamic Thermal Behaviour of Power Cables", was accomplished at the University of Gent. From February 2013 he is a PhD Candidate at the University of Mons, working in the Framework of the GREDOR project, a project aiming to increase the capacity of renewable generation in the Walloon Energy System. His work is focused on the stochastic phenomena such as wind power and load prediction.


Construction of aggregated scenario trees for the day-ahead prediction of wind power


Electricity grids throughout the world evolve rapidly due to the integration of Renewable Energy Sources. Although wind energy contributes το a high extent to this emergence of green power generation, the stochastic nature of wind raises difficulties concerning the management of electrical networks. In this paper, we establish a scenario tree which will assist system actors (Transmission System Operators , Distribution System Operators, providers) during the day-ahead management of the grid (electricity market aspects, reserve allocation,...). The impact of the length of historical data and of the implemented prediction models on the volatility of the simulated scenario trees will then be investigated.


A prediction model is first built based on ARMA time series so as to produce a robust day-ahead wind speed forecast. Comparing this predicted evolution with the real one, a model for the forecast error is then computed. The two models are thereafter aggregated and simulated in order to construct a fan of scenarios. This fan is reduced till a limited number of representative scenarios is achieved and then transformed into a multistage tree. The impact of the implemented prediction model and of the considered historical data is then evaluated by comparing the volatility of the simulated scenarios.

Main body of abstract

In this work, time intervals of 15 minutes are considered, for the measurements as well as for the predicted values. Two ARMA series are employed; one for the prediction of the wind speed, and another one for the prediction of the error on the wind speed forecast. The latter is defined by comparing the predicted wind speed values to the measured ones. Note that the parameters of the ARMA models are chosen based on the Akaike Information Criterion (AIC). Prediction and error are then combined in a Monte Carlo process, and a large fan of scenarios with equal probabilities of occurrence is achieved.

Typically, these scenarios are employed to assist system actors during the day-ahead management of the grid. The purpose is to discretize the searching space related to the underlying stochastic optimization process. A high number of scenarios will increase the richness of the solution, at the expense of the computational burden, so that a trade-off has to be found. In this work, a reduction technique based on the Kantorovich distance is applied to the fan of scenarios. Once it is reduced to a desirable size, the fan is transformed into a multistage tree by deleting inner nodes without changing the number of leaves. Finally, by considering typical day-ahead management problems that are encountered in the Belgian medium voltage grid, we show how a scenario tree which integrates the prediction error is superior to a tree that does not take into account that error.


In this work, based on algorithms used for the prediction of the wind power generation, a reduced multistage scenario tree is constructed, so as to be used in day-ahead management tools (day-ahead closure of the electrical market, reserve allocation,…). The presented method is based on the results of the European project named WILMAR. Here, compared to the bibliography, the impact of the historical data length and of the considered prediction models is evaluated. Moreover, in the future, the reduced wind scenario tree will be completed by the additional consideration of solar power generation.

Learning objectives
This work's main objective is to define a way in which stochastic issues, like wind power generation, could be approached by a reduced scenario tree enabling to take all the appropriate control actions for the day-ahead maintenance of the electrical grid. Moreover, the influence of various parameters on the volatility of the simulated scenarios, such as the length of the historical data or the coefficients of the prediction models, is tested with the estabished tool.