The suboptimal environment of wind farms poses significant challenges for maintenance and results in the generation of complex abnormal data in the transmitted signals. This issue is particularly severe due to the accumulation of large volumes of abnormal data caused by wind sensor failures and power limitations. In our work, we analyze the mechanisms behind various types of abnormal data (anomalies) generation and outline the structure of the proposed solution. To detect anomalies and reconstruct wind speed-power data, our approach integrates iterative curve fitting for outlier removal, linear interpolation, and density clustering. Tailored strategies are developed to address different types of abnormal data, and the effectiveness of the proposed anomaly detection method is validated through real-world case studies.