Abstract:

Aiming at the problem that unknown samples cannot be effectively distinguished in existing circuit breaker fault diagnosis research, a circuit breaker fault diagnosis algorithm based on the convolutional prototype network is proposed. Firstly, the classification function is constructed using the clustering approach and the probability space is divided based on the distance constraint of the prototype sample point feature space for various types of faults, which can achieve the recognition of a sample set containing unknown fault classes. At the same time, with each type of prototype sample points as the cluster center, the sample feature space distance is used as the optimization target of the convolution feature self-extraction network, which can effectively improve the intra-class aggregation and inter-class dispersion of fault sample features and improve the classification accuracy of the model. Finally, the validity and accuracy of the proposed algorithm are verified based on the field experiment data of 110 kV circuit breaker. The results show that the proposed algorithm can accurately distinguish the unknown faults in the test samples and effectively improve the spatial distribution of fault sample features.

Key words: circuit breaker, fault diagnosis, prototype convolutional network, clustering, unknown class, intelligent maintenance