Image-Based Wind Power Curve Data Cleaning Algorithm via the Matching of Deformation Template

Abstract

The Supervisory Control And Data Acquisition (SCADA) system can collect the wind turbine data, which plays an important role in wind farm maintenances and operations. Due to some unforeseen circumstances, the wind power curve (WPC) contains many abnormal data which should be cleaned. This paper proposed an image-based WPC data cleaning algorithm via the matching of deformation template (MDT). The proposed method can be summarized into the following steps: Initially, the unprocessed data are cleaned by the Local Outlier Factor (LOF) algorithm in three-dimension space (wind speed, rotor speed, wind power). Then, the pre-processed WPC data is transformed into a WPC image with a mapping between the data and the pixels. The reference WPC is aligned to the WPC image via affine transformation based on the bounding box of the contour. In the third step, a feature image representing the confidence of being normal is built based on the distances in eight directions. In this way, the data cleaning task is transformed into an image segmentation problem. The image thresholding method is employed and the optimal threshold t <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">o</sub> is determined by the dissimilarity of Hu moment with the deformed reference WPC. Comparative experiments were operated with some typical data-based and image-based algorithms on the datasets from 17 wind turbines to verify the effectiveness and versatility of the proposed method.