An Anomaly Detection Algorithm for the Converter Valves in HVDC Based on Visible and Infrared Image Fusion

The converter valve is the most core equipment of the HVDC converter system. In recent years, converter stations have gradually started installing visible and infrared inspection equipment in valve halls, using multi-source information to monitor the flexible DC converter valves. This paper proposes a novel anomaly detection algorithm based on the improved YOLOv7 and fusion of infrared and visible images, which is used to detect temperature rise anomalies. Firstly, two features extraction paths, visible image and infrared, are added to the backbone network to extract visible and infrared features, respectively. Then, train the proposed algorithm by generating temperature rise anomalies. Finally, the effectiveness of this method were verified through various experiments. The results show that the anomaly detection method can achieve a good accuracy to abnormal temperature rise detection of the converter valves.