Shuntless current measurement in sector-shaped power grid cables: A simulation and machine learning approach

Abstract

This work introduces a calibration-free and retrofittable contactless current measurement method for three-phase segmented power cables, combining magnetic field sensing with machine learning-based inverse modeling. Unlike traditional approaches, the proposed technique does not require prior knowledge of conductor positions or a closed-loop sensor geometry, making it suitable for seamless integration into existing infrastructure. A finite element simulation framework was developed to generate realistic magnetic field data for cables with arbitrary cross-sections. These data were used to train various lightweight machine learning models, including K-Nearest Neighbors (KNN), Least Squares Support Vector Machine (LSSVM), and tree-based ensembles (ExtraTrees, CatBoost, XGBoost, LightGBM, HistGradientBoosting). The method was validated on a custom-built experimental test bench using a real segmented cable. Under symmetric current conditions, a best-case RMSE of 0.086 A was achieved from real measurements, corresponding to a current reconstruction error of 1.99% for 2 A input currents. Robustness was further demonstrated by evaluating the model under simulated perturbations such as sensor displacement, current imbalance, and measurement noise. The results show a strong generalization from simulation to reality without retraining and a low sensitivity to mechanical or electrical distortions. These findings confirm the potential of the method for integration into smart grid systems and resource-constrained embedded environments, offering accurate, scalable, and calibration-free current sensing. Compared to existing contactless current measurement techniques, this approach combines physical interpretability with data-driven adaptability, enabling accurate sensing even in segmented or partially accessible cables without calibration or closed-loop geometries.