Early detection of arc faults in DC microgrids using wavelet-based feature extraction and deep learning

This work presents an approach for anomaly detection using autoencoders and wavelets to identify arc faults in a DC power system, where Cassie arc model is used for synthetic arc fault generation. The system uses a deep learning technique called an autoencoder to detect anomalies in the signal. The autoencoder is trained on normal, fault-free data. It can then detect faults by identifying deviations from the normal data. The work compares the effectiveness of using raw data versus wavelet-filtered data for training the autoencoder. The results show that wavelet-filtered data leads to better performance. In one test, the autoencoder using wavelet-filtered data achieved a 97.52% probability of detecting faults, while the autoencoder using raw data achieved only a 57.85% probability. The results demonstrated that wavelet-filtered data can significantly improve the performance of autoencoder-based anomaly detection.