Multiple domain identification of fault arc based on KPCA-LSTM method

Abstract

Arc faults induce multi-domain variations, leading to low accuracy in identifying multi-domain arc faults. To address this issue, a multi-domain arc fault identification method based on Kernel Principal Component Analysis-Long Short-Term Memory (KPCA-LSTM) is proposed. This method utilizes Kernel Principal Component Analysis (KPCA) to reduce the dimensionality of multi-domain arc fault features, obtaining principal component vectors. Long Short-Term Memory (LSTM) is applied to extract features from the reduced dimensions, combined with Discrete Wavelet Transform (DWT) to model and extract frequency-domain features of arc faults. Furthermore, to mitigate noise interference, a signal threshold denoising method based on wavelet modulus maxima theory is proposed. The detail coefficients are calculated based on the type of arc fault point to capture signals across different frequency bands for multi-domain arc fault identification. Experimental results demonstrate that KPCA performs optimally in dimensionality reduction, achieving high model training accuracy. The accuracy of identifying individual branches and different types of faults exceeds 98 %, surpassing the Support Vector Machine (SVM) method. KPCA-LSTM exhibits superior performance in transient and continuous breakpoint faults, effectively improving the accuracy and efficiency of arc fault identification in power systems, thereby providing strong support for the safe operation of power systems.