An Explainable Physics-Informed Learning Framework For Single-Ended Fault Location In 100% Inverter-Based Grids

Inverter-dominated transmission grids challenge conventional protection because converter control and current limiting distort fault transients and weaken phasor-domain fault signatures. This paper addresses the unresolved problem of accurate and unambiguous single-ended fault location in 100% inverter-based systems, particularly when equidistant faults produce identical propagation distances from one terminal. In this work, introduces a traveling wave (TW)-based, explainable physics-informed machine learning (PIML) framework for single-ended fault distance estimation in 100% inverter-based transmission systems comprising both grid-forming (GFM) and grid-following units. The method extracts high-frequency features from measured fault-induced current TWs using a high-order synchrosqueezing transform and embeds them into a PIML model constrained by the telegrapher’s equations. To resolve cases where multiple fault points yield identical propagation distances from a single terminal, a minimal-sensor observability mechanism is proposed to uniquely identify each location using combined distance signatures from selected GFM buses. Shapley additive explanations ensure transparency of feature influence on the predicted fault distance. Electromagnetic transient-grade simulation in PSCAD/EMTDC on the IEEE 9-bus inverter-dominated system demonstrates reliable discrimination of close-in and equidistant faults and reduces the test mean squared error by approximately 74–88% relative to three state-of-the-art data-driven baselines under diverse inverter controls and fault conditions. Overall, the proposed framework provides a transparent and sensor-efficient route toward protection-grade fault location in fully inverter-based grids.