CNN-LSTM neural network with attention mechanism for fault diagnosis of high voltage circuit breakers

Abstract

In the current fault diagnosis of high-voltage circuit breakers (HVCB), the traditional methods are limited by prior knowledge and manual experience. The accuracy of convolutional neural networks (CNN) isn't good enough, and there are often false positives or false negatives, which don't realize the effective monitoring of HVCBs. To solve this problem, this paper proposes a fault diagnosis method based on convolutional neural network-long short-term memory network (CNN-LSTM) with attention mechanism. First, the CNN with attention mechanism is used to extract the features of the mechanical fault. The fusion of the attention mechanism makes the CNN pay more attention to the high-importance features, which solves the problem that the CNN does not distinguish the importance of the features and causes the loss of important features. Then, the long short-term memory (LSTM) module is introduced as a classifier to improve the ability to utilize and process the features extracted by the CNN module, so as to complete the task of fault classification. The experimental results show that, compared with CNN, CNN-LSTM and other intelligent diagnosis methods, the CNN-LSTM with attention mechanism proposed in this paper obtains higher-precision diagnosis results, which provides a reliable basis for fault diagnosis of HVCBs.