Anomaly data identification for wind farms based on composite machine learning

Abstract

The harshness and uncertainty of the operating environment have caused a large amount of anomaly data to wind farms, so clean and valid operation data are essential for smart wind power operation and maintenance. Therefore, this paper proposes a composite machine learning algorithm based on the horizontal vertical quartile method and extreme learning machine (ELM) to recognize anomaly wind speed-power data in the wind farm. First, the anomaly points of the wind speed-power data are identified from a bilateral relationship of wind speed and power output by using the horizontal and vertical quartile methods. Second, the effects of different quartile methods on the cleaning effect are compared, and the optimal method with a combination of horizontal and vertical quartiles is selected to identify the abnormality of the wind speed-power data in wind farms. Then, the wind speed-power data could be tagged after anomaly identification, and the sample library is setup. After that, the ELM is trained to learn the data features, and it could be applied to anomaly data identification in the future operation of wind farms. Finally, an example is used to verify the effectiveness of the proposed method.