Insulator Self-explosion Fault Detection Based on Transfer Learning
Download as PDF
DOI: 10.25236/mmmce.2020.034
Author(s)
Huohua Li, Xiaogang Fu, Tairu Huang
Corresponding Author
Huohua Li
Abstract
Insulator is an important device used to support wire and electrical isolation in power system. It is of great significance to realize intelligent detection of insulator state for stable and reliable operation of transmission line. Considering the randomness of insulator image state and the complexity of image background, as well as the small scale of insulator fault dataset collected at present, the classic AlexNet network structure is built, and the trained model is in the state of under fitting, which cannot meet the requirements of detection accuracy. In order to solve this problem, the transfer learning method is introduced and the fine tuning strategy of the pre-training network is elaborated. In the experiment, based on the pre training network of vgg16, some weight parameters of the vgg16 network are frozen, and the self-defined full connected layer is trained. The convergence speed and detection accuracy of the new model in the test set and verification set are effectively improved.
Keywords
Fault detection, AlexNet, transfor learning, pre-training network, vgg16