A photovoltaic anomaly data identification method based on image feature detection

The supervisory control and data acquisition （SCADA） system of photovoltaic （PV） power plants records a substantial volume of operational and maintenance data that is crucial for the routine maintenance. However， abnormal data resulting from extreme weather， sensor failures， and other factors significantly degrade data quality， thus affecting PV power forecasting and routine maintenance. To address this， this paper introduces an anomaly data identification algorithm based on image feature detection and dual-threshold processing. This method maps numerical data to images， transforming the anomaly detection problem into an image processing problem. First， abnormal data is categorized into negative value anomalies， discrete anomalies， and stacked anomalies. Discrete anomalies are identified based on the density of image data. Next， stacked anomalies are detected using Canny edge detection and Hough transform， with a dual threshold image processing mechanism introduced to enhance the method’s generalizability. Finally， the proposed method is compared with traditional statistical methods using a real-world dataset， demonstrating its adaptability.