Short-Term Wind Power Forecasting
Published in João P. S. Catalão, Electric Power Systems, 2017
Gregor Giebel, Michael Denhard
Bremnes [114] developed a probabilistic forecasting technique, estimating the different quantiles of the distribution directly. In another study [115], Bremnes described his method of local quantile regression (LQR) in more detail, and showed that, for a test case in Norway, HIRLAM forecasts have a lower interquantile range than climatology, which means that the HIRLAM forecasts actually exhibit skill. LQR HIRLAM features about 10% better in economic terms than pure HIRLAM forecasts, increasing the revenue from ca.75%–79% of the ideal income (without any forecast errors) to ca. 79%–86%, depending on the horizon. However, his pure HIRLAM forecasting did not have an upscaling or MOS step, so this might have worked in favor of LQR in comparison. Bremnes proposed to use the method to reduce the large amount of information found in meteorological ensembles. The motivation for this was that he could show that the economically optimal quantile was not the central (“best”) quantile, but one given by the relative prices of up- and down-regulation.