Vision Based State Recognition of 220kv Disconnect Switches in Power Substations

Abstract

Remotely monitoring and controlling power substations are popular nowadays. For such systems, it is essential to continuously keep track of devices' conditions. This paper presents a vision-based method for recognizing states, open and closed, of 220kV disconnect switches in power substations. Images of size 512×512 were collected at a 220kV power substation for training and test whereas the EfficientDet model D0 was utilized for transfer learning. Experiment results show that the model can accurately recognize states of 220kV disconnect switches with mAP of 99.74%.