Wind Farm Power Forecasting for Less Than an Hour Using Multi Dimensional Models

This research focuses on prediction of wind farm power for horizons of 0-10 minutes and not more than one hour using statistical methods. These short term predictions are relevant for both transmission system operators, wind farm operators and traders. Previous research indicates that for short time horizons the persistence method performs as well as more complex methods. However, these results are based on accumulated power for an entire wind farm. The contribution in this paper is to develop multi-dimensional linear methods based on measurements of power or wind speed from individual wind turbine in a wind farm. These multi-dimensional methods are compared with the persistence method using real 1 minute average data from the Sheringham Shoal wind farm with 88 turbines. The results show that the use of measurements from individual turbines reduce the prediction errors 5-10% and also improves the prediction error variance estimate compared to the persistence method. We also present convincing examples showing that the predictions follow the wind farm power over a window of an hour.