Final Report - Forecasting Exercise
 

 supported by the PROFIT program by the Ministerio de Educación y Ciencia.

La The Spanish Wind Energy Association (AEE) has promoted among its technical activities one project related with the prediction of wind farm energy production, known as the Prediction Exercise. This project has been financed by the Spanish Wind Energy Association AEE, the owners of the wind farms and a subvention by the PROFIT programme of the Ministry of Education. Different prediction companies have made daily operative predictions for seven wind farms for a period lasting from august 2004 till end of march 2006.

• The outcomes of the different prediction models have shown the importance of the quality of the input global meteorological data, which must improve in both, resolution of the grid and combined use of different mesoscale models.
• It is also very important to reduce the time between the recollection of meteorological data, its treatment and the output of the meteorological centres, especially for the intradaily markets.
• To avoid divergence between the results from the various models, in order to calculate the error, two parameters are used:
• EMAP: Mean Absolute Prediction Error for a certain period. It can be interpreted as the percentage that the addition of all absolute errors represent over the total production of a period.
• EMAE: Mean Absolute State Error, the same as before, but divided by the nominal power of the wind farm, also for a certain period. It can be interpreted as the mean value of the absolute error of the predicted capacity factor.

• For this exercise it has been revealed that statistical models show an acceptable accuracy at a reasonable price.
• The mean EMAP for all predictors is around 45%
• The lowest monthly EMAP that ca be reached is about 25% and further improvements may only be achieved by improvements in the input meteorological data, the supply of live data from the wind farms and the ensemble of models.
• There has not been discovered a relation between the outcomes of the models, its complexity, cost and input data needs.
  The influence of terrain complexity is, either negligible or too small compared with other parameters, so as not to be determining for the accuracy of prediction models
• The grouping of predictions for different wind farms reduces the errors. The maximum reduction is reached by grouping all seven wind farms inb the exercise and is about one half of the original error.
• Short time errors are lower than day-ahead ones, considering for each intraday forecast only its own exclusive period. In any case, results are not conclusive because of lack of meteorological data and data flow form the wind farms in real time.
• Changes in determining the economical cost of prediction deviations have had a big impact on the application of the models and will be further so in the future with the application of different criteria depending on the errors being for or against the deviation of the system.
   

[+] Ver informe

Forecasting Exercise
 

As electricity generators, and given the considerable penetration levels of an energy source previously considered supplementary, wind power operators are obliged to integrate production within the electricity market in order to optimise spinning reserve.In this sense, the Royal Decree (Real Decreto) 436/2004 marks an important step.

The regulation enables wind power to trade on the electricity market, both in the case of production supplied to distributors and in that of entering directly through a trading agent. This market participation involves at least two key components: organising combined offers with other special regime generators and, in order to make offers, forecasting wind plant production. Regarding forecasting, AEE launched, in mid-2004, its Forecasting Exercise (Ejercicio de Predicción).

This is a pioneering project in the global context given that it is coordinated and executed by the wind operators themselves as part of an unerring commitment to optimising their participation in system operations.

The project’s main objectives are to evaluate the maximum attainable results from forecasting with currently available tools and to analyse the influence of different factors on forecasting error. Such factors mainly involve orography, thermal winds, forecast model type and programming timelines, etc.

INSTITUTIONAL COLLABORATION
The forecasting exercise includes a market simulation component, carried out with the collaboration of the electricity market operator—Operador del Mercado de la Electricidad (OMEL). The sector has received enthusiastic support for the initiative from OMEL, whose refreshing professionalism and thoroughness— qualities difficult to find in other countries—have been greatly appreciated. The exercise has also received support from the state energy efficiency agency, Instituto para la Diversificación y el Ahorro Energético (IDAE), as well as from the Ministry of Education and Science’s PROFIT programme. Very soon, the national meteorological institute, Instituto Nacional de Meteorología, will also collaborate by providing broad-based meteorological data.

In short, the exercise is an extremely important advance on the wind forecasting agenda on a global scale and shows the great innovative scope of the wind sector in Spain.

SEVEN WIND PLANTS AND SIX FORECASTING MODELS
AEE’s Exercise for Programming Electrical Production from Wind Power (Ejercicio de Programación de la Producción Eléctrica de Origen Eólico)—better known as the Forecasting Exercise—analyses the results of six forecasting models applied across seven wind plants. The plants in question were selected as a cross-section of Spanish wind installations. The exercise is a pioneering one in a global context as it is the first one devised, executed and led by the sector itself. Previously, such programmes have been the domain of the electricity system operators, and have always been restricted to a sole forecasting model. AEE’s forecasting exercise, planned towards the end of 2003, was launched on 14 April 2004.

The first step was to adapt the various forecasting models to the different wind plant sites. As a pioneering project with few precedents, diverse methods and approaches existed, regarding both the existing wind forecasting models and the basic data criteria of each one. This complicated the result comparisons. One illustration of this is the fact that there was not even a common and accepted definition of forecasting error. After sorting out basic forecasting criteria and establishing the general bases of participation for the companies involved, six companies (Meteológica, Meteotemp, CENER, Casandra, Garrad & Hassan and Meteosim) agreed to submit their wind forecasting models.

The models were applied across seven wind plants: two in Castile and Leon (Páramo de Poza and Villacastín); one in Galicia (Pena da Loba); one in Aragon (El Pilar); one in Castile La Mancha (Muela); one in Andalusia (Buenavista) and one in the Canary Islands (Punta Gaviota). These plants belong, respectively, to: Enerfin (Elecnor group), Genesa (utility Hidrocantábrico); ECYR (utility Endesa); utility Gas Natural; utility Iberdrola, Enerfin and IDAE. More recently, Aleasoft has joined the exercise with short term forecasting. The seven wind plants were selected as a cross-section in terms both of turbine technology and site topography (coastal, plane,mountainous, etc.).

The aim here is to obtain results that may be accurately extended to the rest of Spain’s wind plants. The exercise will run for a full year in order to enable data compilation over the four seasons.