Wind power is an unstable power source, as its output fluctuates drastically according to the weather. Such instability can cause sudden large-scale changes in output, called ramp events; the frequency of such events is relatively low throughout the year but they could negatively affect the supply-demand balance in a power system. This study focuses on an alerting scheme of wind power ramp events for a transmission system operator to support operational decisions on cold reserve power plants. The ramp alerting scheme is implemented from the viewpoint of supervised learning by using the prediction results of wind power output. In particular, the authors address the class imbalance problem, as the accuracy of ramp event prediction tends to be low because of the infrequency of such ramp events in the database used for learning. In this study, several data sampling strategies are proposed and implemented to overcome the problem in the ramp alert task. The effectiveness of the proposed data sampling framework is evaluated experimentally by predicting real-world wind power ramps, based on a dataset collected in Japan. The experimental results show that the proposed framework effectively improves the ramp alert accuracy by addressing the class imbalance problem.
ASJC Scopus subject areas