Genetic Programming for Wind Power Forecasting and Ramp Detection

Abstract

In order to incorporate large amounts of wind powerinto the electric grid, it is necessary to provide gridoperators with wind power forecasts for the day ahead,especially when managing extreme situations: rapidchanges in power output of a wind farm. These so-calledramp events are complex and difficult to forecast.Hence, they introduce a high risk of instability to thepower grid. Therefore, the development of reliable rampprediction methods is of great interest to gridoperators. Forecasting ramps for the day ahead requireswind power forecasts, which usually involve numericalweather prediction models at very high resolutions.This is resource and time consuming. This paperintroduces a novel approach for short-term wind powerprediction by combining the Weather Research andForecasting advanced Research WRF model (WRF-ARW) withgenetic programming. The latter is used for the finaldownscaling step and as a prediction technique,estimating the total hourly power output for the dayahead at a wind farm located in Galicia, Spain. Theaccuracy of the predictions is above 85 percent of thetotal power capacity of the wind farm, which iscomparable to computationally more expensivestate-of-the-art methods. Finally, a ramp detectionalgorithm is applied to the power forecast to identifythe time and magnitude of possible ramp events. Theproposed method clearly outperformed existing rampprediction approaches.