Forecast of Infrequent Wind Power Ramps Based on Data Sampling Strategy

The introduction of wind power generation has been promoted in Japan. However, wind power is an unstable power source because its output varies according to the weather. Particularly, sudden changes in output, which could adversely affect the power system, are called ramps and may cause serious problems in the power system. In this paper, the authors discuss the ramp event forecast by using classifiers. A serious issue in this setup is that classification based forecast tends to be inaccurate since the occurrence of such a ramp is relatively rare. This problem is called the class imbalance problem in the machine learning field. To overcome the class imbalance problem in ramp forecast, several data sampling approaches are implemented. The effectiveness of these sampling approaches is experimentally evaluated by using a real-world wind power generation dataset. The results show that the implemented approaches drastically improved the forecast accuracy.