Support Vector Regression-based Cable Temperature Prediction for Digital Twin Applications

The development of new-type power systems has positioned temperature prediction of power cables as a critical factor in ensuring power supply reliability. Conventional computational methods for cable temperature estimation struggle to meet the real-time requirements of Digital Twin applications. This study proposes a Digital Twin-based temperature prediction methodology that integrates machine learning with finite element modeling. Focusing on 10 kV three-core cross-linked polyethylene (XLPE) cables, we establish a two-dimensional finite element model incorporating electromagnetic-thermal field coupling to generate fundamental temperature datasets. A six-dimensional feature vector is selected to construct a Support Vector Regression (SVR) prediction model, enhanced through radial basis function kernel implementation and Bayesian-optimized hyperparameter tuning. The optimal parameter combination ($\mathbf{C}=\mathbf{1 0}, \gamma=\mathbf{0. 2 5}$) is determined via five-fold cross-validation, significantly improving model generalization capabilities. Experimental results demonstrate that the optimized SVR model achieves a prediction accuracy of 0.995 with a root mean square error of 0.2. The computational time is reduced to 0.05 seconds, representing a 200-fold efficiency improvement over conventional finite element methods. This approach enables rapid and precise prediction of cable temperature distribution through real-time cyber-physical mapping, effectively validating the synergistic potential of machine learning and finite element integration for cable temperature prediction in Digital Twin applications.