Application of change-point analysis to abnormal wind power data detection

The abnormal data of wind power could be caused by many on-site situations, such as meteorological conditions, control strategies and communication environments, which must be detected before put into work. This paper presents data detection methods by taking the abnormal data as the change points in wind power, which means an unknown moment when there are some changes appearing to the studied system abruptly. Due to the fluctuate nature and auto-regressive features of wind power at different time scales, change-point analysis in this paper is discussed on perspectives of cumulative probability distribution, changes on regression modeling characteristics and hypothesis testing about influences brought by some special key factors. Historical data of a wind farm in western China are studied to explain and verify the proposed detection methods, which could help catch the abnormal moment or period of wind power effectively and clarify the causes of such data.