Frequency-based Multi Task learning With Attention Mechanism for Fault Detection In Power Systems

The prompt and accurate detection of faults and abnormalities in electric\ntransmission lines is a critical challenge in smart grid systems. Existing\nmethods mostly rely on model-based approaches, which may not capture all the\naspects of these complex temporal series. Recently, the availability of data\nsets collected using advanced metering devices, such as Micro-Phasor\nMeasurement units ($\\mu$ PMU), which provide measurements at microsecond\ntimescale, boosted the development of data-driven methodologies. In this paper,\nwe introduce a novel deep learning-based approach for fault detection and test\nit on a real data set, namely, the Kaggle platform for a partial discharge\ndetection task. Our solution adopts a Long-Short Term Memory architecture with\nattention mechanism to extract time series features, and uses a\n1D-Convolutional Neural Network structure to exploit frequency information of\nthe signal for prediction. Additionally, we propose an unsupervised method to\ncluster signals based on their frequency components, and apply multi task\nlearning on different clusters. The method we propose outperforms the winner\nsolutions in the Kaggle competition and other state of the art methods in many\nperformance metrics, and improves the interpretability of analysis.\n