Abstract: To address the challenges of complex fault characteristics in distribution networks caused by high penetration of distributed generation (DG) and the insufficient tolerance of traditional fault location methods to high-impedance and communication disturbances, a small-current grounding fault location strategy based on multi-terminal synchronous measurement and graph convolutional networks is proposed. This method constructs an inverter-type DG Norton equivalent model considering the dynamic characteristics of phase-locked loops to quantify the impact of weak-fed current on sequence network parameters. It employs IEEE 1588v2 transparent clock technology combined with sliding window time-stamp correction to achieve microsecond-level hard synchronization in harsh communication environments. By extracting transient zero-sequence energy spectra using generalized S-transform and mapping them to graph node attributes after dimensionality reduction via kernel principal component analysis, the strategy inputs a two-layer graph convolutional network to mine topological spatial correlations for fault section identification. Hardware-in-the-loop tests demonstrate that the strategy maintains over 90% location success rate under extreme conditions such as 1000Ω high-impedance grounding and 30% packet loss rate, while meeting the rapid-response requirements of distribution automation with end-to-end delay compliance