Simulation-Driven digital twin framework for fast virtual measurement and optimization of a 71-Level hybrid cascaded H-bridge inverter
Measurement: Journal of the International Measurement Confederation, cilt.286, 2026 (SCI-Expanded, Scopus)
- Yayın Türü: Makale / Tam Makale
- Cilt numarası: 286
- Basım Tarihi: 2026
- Doi Numarası: 10.1016/j.measurement.2026.122377
- Dergi Adı: Measurement: Journal of the International Measurement Confederation
- Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Compendex, INSPEC, Academic Search Ultimate (EBSCO), Engineering Source (EBSCO), Materials Science & Engineering Collection (ProQuest), Technology Collection (ProQuest)
- Anahtar Kelimeler: Hybrid cascaded H-bridge inverter, Multilevel inverters, Noise sensitivity, Pareto optimization, Power loss prediction, Simulation-driven digital twin, Soft sensing, Total harmonic distortion, Virtual measurement, XGBoost
- Bursa Uludağ Üniversitesi Adresli: Evet
Özet
High-level multilevel inverters, such as the 71-level hybrid cascaded H-bridge topology, can provide high waveform quality, but their design and optimization require time-consuming switching-level simulations. This paper proposes a simulation-driven digital twin framework based on XGBoost for fast virtual measurement of key inverter performance indicators. A MATLAB/Simulink dataset with 700 valid operating points was generated by varying the modulation index, load resistance, and PWM strategy. The trained model estimates total harmonic distortion and inverter power loss with R2 values above 0.9999 on validation and test data, while switching-event information is used as an additional operating indicator in the optimization layer. The corrected prediction and residual analyses show close agreement between the digital-twin estimates and the simulation reference values. The surrogate model reduces the inference time to 0.0437 ms per sample, corresponding to a speedup of approximately 686,430 times compared with a 30 s high-fidelity simulation, with a memory footprint of 318.23 KB. A noise-sensitivity test also shows that the model is highly accurate under nominal conditions but sensitive to perturbed sensor-related inputs, indicating the need for filtering or uncertainty-aware preprocessing before practical sensor-based deployment. Finally, the digital twin is used for Pareto-based operating-point selection, showing how harmonic distortion, switching activity, and power loss can be traded off for application-specific control decisions. The proposed framework therefore provides a fast soft-sensing and optimization layer for simulation-based monitoring and design of high-level multilevel inverters.