Abstract-An increasing number of utilities participating in the energy market require short term (i.e. up to 48 hours) power forecasts for renewable generation in order to optimize technical and financial performances. As a result, a large number of forecast providers now operate in the marketplace, each using different methods and offering a wide range of services. This paper assesses five different day-ahead wind power forecasts generated by various service providers currently operating in the market, and compares their performance against the stateof-the-art of short-term wind power forecasting. The work focuses on how power curve estimations can introduce systematic errors that affect overall forecast performance. The results of the study highlight the importance of: accurately modelling the wind speedto-power output relationships at higher wind speeds; taking account of power curve trends when training models; and the need to incorporate long-term (months to years) power curve variability into the forecast updating process.Index Terms-wind power forecast; wind turbine power curve; short-term forecasting; forecast assessment; wind energy.

No abstract

The evaluation of wind energy forecasts is a key task for those involved in the wind power sector, and the accurate evaluation of forecasts is fundamental to make informed decisions both in business and research. To evaluate the accuracy of a forecast, observed values must be compared against forecast values over a test period. At times, however, the actual generation of a wind farm can be affected by factors that are outside the scope of the forecast model. Evaluating a forecast using a data set that includes such out-of-scope observations might give a biased or inconsistent assessment. In the data preparation phase, then, the evaluator should identify out-of-scope data and decide whether to include or remove these from the data set. In this paper, we carry out an empirical study based on data from an existing wind farm and a number of day-ahead forecasts in order to highlight the effects of including in-and out-of-scope data on forecast accuracies. The results show that the outcome of the evaluation varies significantly depending on the criteria adopted in the data selection.

No abstract