Digital Twin for Parameter Monitoring of Modular Multilevel Converters in HVDC Transmission of Offshore Wind Energy

Modular multilevel converters (MMCs) are critical to modern power grids, necessitating robust prognostic health management strategies. This paper proposes a novel, data-driven digital twin framework using a temporally decoupled tri-model architecture that employs three feed-forward neural networks for real-time MMC health monitoring and fault diagnostics. This framework is made of a performance digital twin model that emulates converter dynamics and provides a healthy baseline for anomaly detection. The second digital twin is a diagnostic classification model that identifies healthy, DC-side short circuit, AC-side open circuit, and AC-side short circuit fault conditions. The third is a prognosis model that estimates arm inductance degradation using a novel approach of utilizing short-time Fourier transform-based feature extraction from arm voltage and arm current signals. The architecture was validated under extensive test data, including steady-state, AC/DC side faults, and inductor degradation down to 50% of nominal value. The performance model achieved a root square error (RMSE) of 0.0075, the classification model reached accuracy above 0.9 for key faults, and the prognosis model attained an average RMSE of 0.085 in inductance prediction. The proposed DT framework offers an accurate, and computationally efficient approach to MMC prognostics, providing essential inputs for estimation of remaining useful life estimation and enabling condition-based maintenance.