“…Traditional data-driven methods for fault localization, such as travelling-wave [Parsi et al, 2020] and impedance based ones [Aucoin and Jones, 1996], require high grid observability and sampling rates that are technically challenging and expensive for bulky systems [Sundararajan et al, 2019] or even known distribution of renewables [Owen et al, 2019. Another line of algorithms leverages deep neural networks capabilities [Li et al, 2019, Li and Deka, 2021a, Zhang et al, 2020a, Misyris et al, 2020; however, these methods suffer from high requirements on the amount of phasor-measurement unit data. The latter lead to inability to make a accurate and timely detection in time-changing environment that is intrinsic for extreme weather events and, therefore, compromises power grid security.…”