Abstract: Poor contact is a common cause of fire in low-voltage lines. When the poor contact reaches a certain high temperature, it will lead to glowing contact fault of electrical lines, which has great potential fire hazards. In order to diagnose the glowing contact fault of electrical circuit, this paper proposes a time-frequency domain feature extraction method based on empirical wavelet transform (EWT), and uses the whale optimization algorithm (WOA) to obtain the optimal input weights and hidden thresholds of extreme learning machine (ELM) based on the fusion feature set, and proposes a neural network model based on EWT-WOA-ELM. The results show that the optimal number of neurons in the hidden layer of the model is 18. The cross validation average accuracy and average cross entropy loss of this model are 96% and 0.623 9, respectively, achieving fault diagnosis for normal state, early and late stages of glowing contact under different operating conditions. Different laboratory data were used to validate this model, and it was found that the model's recognition state was consistent with the actual state.

Key words: glowing contact, fault diagnosis, machine learning, extreme learning machine