Metallic sheath connection errors in high-voltage cables represent a typical defect in high-voltage cable sheath grounding systems. For cross-bonded grounding boxes

there are 36 possible connection types. Conventional sheath current measurement methods can only obtain the magnitude of the current

making it difficult to accurately identify cable sheath connection configurations. we developed equivalent circuit models of a three-phase nine-section cross-bonded high-voltage cable under both correct and incorrect sheath bonding conditions. A method based on graph neural networks (GNN) was proposed

utilizing the sheath currents at both ends of the three loops to identify 36 possible cross-bonding configurations. Moreover

a 220 kV cable system was taken as a case for simulation. The results demonstrate that the magnitudes and phasors of sheath currents at both ends of the three loops can effectively distinguish not only between correct and incorrect cross-bonding states

but also among different types of connection errors. The proposed GNN-based classification model for high-voltage cable cross-bonding configurations can accurately identify 36 cross-bonding sheath connection scenarios

achieving an overall classification accuracy exceeding 92%. The five-fold cross-validation results fully verify the effectiveness and superiority of the proposed model in the identification of high-voltage cable cross-bonding connection configurations.