| Title |
Analysis of Prony and Matrix Pencil Methods for Online Power Grid Oscillation Analysis |
| Authors |
손민환(Min-hwan Son) ; 송성윤(Sung-yoon Song) |
| DOI |
https://doi.org/10.5370/KIEE.2026.75.8.1721 |
| Keywords |
Prony Analysis; Matrix Pencil; Randomized SVD; Low-Frequency Oscillation; Modal Identification; Online Monitoring |
| Abstract |
Low-frequency oscillations (LFO) in inverter-based renewable energy systems present challenges for online grid monitoring. This paper compares three modal identification techniques?Prony analysis, full SVD-based Matrix Pencil (MP-SVD), and randomized SVD-based Matrix Pencil (MP-RSVD)?for identifying damping ratios and frequencies from time-domain voltage signals. Synthetic benchmark signals are constructed with reference to the 0.6 Hz local-oscillation frequency band reported in the April 2025 Iberian blackout?not the actual event recordings?and are analyzed across five noise conditions (SNR ∞, 40, 30, 20, 10dB). Results show that Prony analysis achieves the lowest RMSE in low-noise conditions when order is properly tuned, but exhibits high sensitivity to order selection and numerical instability at high orders. MP-SVD and MP-RSVD show comparable accuracy in low-to-moderate noise, while MP-RSVD reduces computation time by 2.0?2.7× through randomized projection. At SNR 10dB, MP-SVD experiences rank inflation leading to 3.68× higher RMSE, whereas MP-RSVD maintains stable performance through target rank constraints. These results suggest MP-RSVD offers a practical accuracy?computation trade-off for online grid monitoring, though Prony analysis remains advantageous for offline tuned applications. |