Cold ironing is increasingly adopted in ports to reduce local emissions by powering docked vessels through shore-based electrical systems. However, its deployment introduces protection and monitoring challenges associated with the medium-voltage equipment, grounding constraints, renewable integration, and the limited availability of validated fault datasets. High-impedance faults (HIFs) are particularly difficult to detect because their low and irregular currents can appear very similar to normal operating conditions, allowing them to escape conventional protection devices. This paper proposes an unsupervised HIF detection framework for a cold-ironing microgrid using a Temporal Convolutional Network (TCN) autoencoder trained only on non-fault operating data. To enhance detection capability, a physics-informed regularization term is incorporated into the loss function, which encourages physically consistent reconstructions by penalizing zero-sequence imbalance. A synthetic dataset was generated in MATLAB/Simscape using a modified IEEE 9-bus microgrid configured for shore-to-ship operation, featuring renewable distributed energy resources and a stochastic arcing fault model. The proposed method was evaluated against a 59N residual-voltage indicator, Isolation Forest (IF), One-Class Support Vector Machine (OCSVM), and a Long Short-Term Memory (LSTM) autoencoder. The results show that the physics-informed TCN achieves an average segment-level Receiver Operating Characteristic Area Under the Curve (AUC) of 0.9989. This represents an improvement over both the standard TCN (0.9275) and traditional 59N indicator (0.5826), confirming the effectiveness of combining unsupervised sequence reconstruction with zero-sequence regularization for protection-focused monitoring in cold-ironing microgrids.
Unsupervised HIF Detection in a Cold Ironing Microgrid via Zero-Sequence-Regularized TCN / Xhani, E., Khan, M.A., Tanoni, G., Comodi, G., Principi, E.. - (2026), pp. 1-6. (2026 International Conference on Smart Energy Systems and Technologies, SEST 2026 esp 2026) [10.1109/SEST67798.2026.11712393].
Unsupervised HIF Detection in a Cold Ironing Microgrid via Zero-Sequence-Regularized TCN
Xhani E.
;Tanoni G.;Comodi G.;Principi E.
2026-01-01
Abstract
Cold ironing is increasingly adopted in ports to reduce local emissions by powering docked vessels through shore-based electrical systems. However, its deployment introduces protection and monitoring challenges associated with the medium-voltage equipment, grounding constraints, renewable integration, and the limited availability of validated fault datasets. High-impedance faults (HIFs) are particularly difficult to detect because their low and irregular currents can appear very similar to normal operating conditions, allowing them to escape conventional protection devices. This paper proposes an unsupervised HIF detection framework for a cold-ironing microgrid using a Temporal Convolutional Network (TCN) autoencoder trained only on non-fault operating data. To enhance detection capability, a physics-informed regularization term is incorporated into the loss function, which encourages physically consistent reconstructions by penalizing zero-sequence imbalance. A synthetic dataset was generated in MATLAB/Simscape using a modified IEEE 9-bus microgrid configured for shore-to-ship operation, featuring renewable distributed energy resources and a stochastic arcing fault model. The proposed method was evaluated against a 59N residual-voltage indicator, Isolation Forest (IF), One-Class Support Vector Machine (OCSVM), and a Long Short-Term Memory (LSTM) autoencoder. The results show that the physics-informed TCN achieves an average segment-level Receiver Operating Characteristic Area Under the Curve (AUC) of 0.9989. This represents an improvement over both the standard TCN (0.9275) and traditional 59N indicator (0.5826), confirming the effectiveness of combining unsupervised sequence reconstruction with zero-sequence regularization for protection-focused monitoring in cold-ironing microgrids.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


