Abnormal data recognition method for wind turbines based on alpha channel fusion

Although image processing technology plays an advanced role in the field of abnormal detection of Wind Power Curves (WPC), enabling accurate identification of various types of abnormal data, it still faces three major challenges: reliance on manually labeled reference samples, representation of data density through rasterization and distance calculations, and insufficient accuracy in identifying stacked abnormal data. To address these problems, this study proposes a simple and efficient method for identifying and cleaning WPC abnormal data. This method does not rely on manually labeled reference samples and achieves the identification of different types of WPC abnormal data by merely adjusting the values of two parameters. The proposed method first employs an alpha channel fusion mechanism to directly represent data density in continuous space, eliminating the need for rasterization. Secondly, it introduces boundary discretization, sequence smoothing techniques, and a boundary completion strategy, which are used to accurately extract the boundaries of normal and abnormal data. Finally, by integrating the Canny edge detection algorithm and image morphology principles, the method achieves precise identification and cleaning of all WPC abnormal data. The 134 WPC datasets from the 2022 Baidu KDD Cup Competition were used as experimental data in this study. The effectiveness of the proposed method was validated through experimental comparisons with seven models on six representative datasets. Additionally, a simple analysis of wind curtailment in the region was conducted by calculating the wind curtailment rates across the 134 datasets. The data and code of this study are available.