Topological Data Analysis for fault classification on transmission lines

This paper proposes a novel method for fault classification on transmission lines through a hybrid model combining Topological Data Analysis and unsupervised Machine Learning. Through persistent homology, signal topological signatures are extracted from each current’s phase and residual current. The spatial properties of the signatures are then fed to a K-means clustering algorithm for fault classification. The method produces accurate and consistent results across a variety of fault records, even when tested under diverse parameterized faults and noise intensities. To investigate further, the model is applied to field records of the French transmission operator RTE (Réseau de Transport d’Electricité) without any parametrization or prior training. The accuracy reflects the generalization abilities of the approach. • Topological Data Analysis for Transmission Line Fault Classification. • A simple and interpretable model with low resource requirements • Robust, tuning-free model for accurate real-time results on raw current data