Abstract:

Aiming at the problems of insufficient arc fault feature extraction and low detection accuracy,a multi-feature fusion improved dung beetle optimizer (IDBO) optimized fusion of the attention mechanism of convolutional neural network (CNN) and bidirectional long short term memory (BiLSTM) neural network series arc fault detection method is proposed.The current time-domain,frequency-domain,time-frequency domain,and signal autoregressive parameter model features are extracted through an experimental platform.The kernel principal component analysis (KPCA) is used to reduce the dimensionality of the features,and they are fused to obtain the feature vectors as the input vectors for CNN-BiLSTM-Attention.The cubic chaotic mapping,the spiral search strategy,the dynamic weight coefficients,and the gaussian cauchy mutation strategy are introduced to improve the dung beetle optimizer.An improved dung beetle optimizer is used to optimize the hyperparameters of CNN-BiLSTM-Attention for the series arc fault diagnosis.The results show that the proposed method can achieve an accuracy of 97.92% in detecting fault arcs and efficiently identify the series arc faults.

Key words: arc fault, improved dung beetle optimizer, multi-level feature fusion, CNN-BiLSTM-Attention