Parameter estimation of submarine power cables in offshore applications using machine learning-based methods

Monitoring electrical parameters of power transmission systems is essential to ensure reliability and optimal operating conditions. This research presents an accurate methodology for estimation of the sequence parameters of submarine power cables using a data-driven approach based synchrophasor measurements. Contrarily of conventional techniques, the proposed methodology is based on supervised machine learning models trained on realistic simulations, which incorporate the physical and geometric characteristics of the power cable, with its seven propagation modes. In practical conditions, the training dataset takes into account noise patterns using well-established modeling methods for phasor measurements. These patterns include time-correlated and statistically coherent disturbances, which are representative of those typically encountered in systems employing Phasor Measurement Units (PMUs). A detailed statistical investigation was also conducted to characterize the empirical distribution of the input data, supporting model design and validation. Remarkably, the proposed algorithm achieves accurate parameter estimation even under elevated noise conditions, requiring as few as 200 training samples. The maximum observed estimation error was approximately 1%, underscoring the robustness, efficiency, and practical viability of the proposed framework for the electrical characterization of submarine transmission systems. • FEM-based modeling captures detailed geometry of submarine power cables. • Time-domain simulations generate realistic terminal phasor measurements. • Synthetic noise and errors emulate PMU uncertainty in field conditions. • ML methods accurately estimate sequence parameters from corrupted data. • Proposed approach works with one-sided measurements and high noise levels.