Summary

With the growth of the photovoltaic (PV) industry and market demand, anomaly detection services for PV farms have become increasingly important. Traditional surveillance systems usually discover anomalies through manual inspection, which requires a lot of manpower and is prone to human negligence. With the development of machine learning algorithms, machine vision can achieve automatic intelligent processing such as image feature extraction, object detection, and anomaly detection. Therefore, an intelligent system based on machine vision is very necessary to be deployed on PV farms for automatic anomaly detection.

The anomaly detection service leveraging machine vision (ADSLMV) can automatically and accurately detect system-level and device-level anomalies in PV farms based on visible light videos and infrared images. The system-level anomalies include human behaviours and external interferences. The device-level anomalies include micro-cracks, hot spots, etc. The application of this service can improve the safety and reliability of PV farms.

Recommendation ITU-T F.743.41 addresses the essential requirements and the framework for ADSLMV in PV farms. In the reference framework, an anomaly event pre-processing and emergency response (AEPER) layer is defined to deal with emergency abnormal events in PV farms. In the Appendix of this Recommendation, use cases for hot spot detection, soiling detection, and foreign object detection are provided.