An infrared image based state evaluation method for cable incipient faults

Abstract

Temperature measurement is frequently used to determine the state of power cable, which is critical for the safety of urban distribution system. This paper combines deep learning and state evaluation, proposing infrared image state evaluation method for cable based on deep learning. First, the concepts of transfer learning and deconvolution are introduced to design the down-sampling and up-sampling networks of the model respectively. Then combined with the data set expansion method, 4000 infrared images of 10 kV cables and their accessories are used for training. Finally, the infrared images of cables and their accessories are statistically analyzed by the method of temperature gradient. The experimental results show that the identification accuracy rate of the model reaches 95.68%. The temperature gradient is distributed in the range of 0–4.2 under normal conditions, and the temperature gradient is greater than or equal to 4.9 in the case of crimping defects. The statistical analysis results show that this method can effectively realize the status of the cable infrared image identity.