Hybrid use of an observer-based minute-scale power forecast and persistence

Abstract

Abstract Lidar-based minute-scale wind power forecasts are valuable to support grid stability and electricity trading. Current methodologies are able to outperform the benchmark persistence only during transient situations and unstable stratification. So far, methods that extend lidar-based forecasts to observer-based forecasts by embedding turbine operational data are not able to outperform persistence during stable atmospheric conditions either. In this paper we therefore analyse the complementary use of an observer-based power forecast and persistence. To do so, we implemented two hybrid approaches: The first is based on a binary decision algorithm, while the second is weighting the two methods by minimizing a cost function. We evaluated 5-minute-ahead deterministic power forecasts of the hybrid and individual models at an offshore wind farm and found the weighting approach to be most skillful. Further, the data set was extended to represent the atmospheric conditions on site for an entire typical year. The weighting approach outperformed the binary decision algorithm for both the 5-minute sample forecasts and the one year-long data set. We discuss the advantages and disadvantages of the two hybrid models and conclude that the weighting approach is the better choice. Further, it can be concluded that also when evaluating the forecasts over a longer period, in this case one year, the additional use of observer-based forecasts is beneficial compared to solely relying on persistence.