Low-frequency oscillation identification through fractal-ridge-optimized MSST and fitting-error-minimized MP

Low-frequency oscillation (LFO) identification in wide-area power systems is often hindered by strong noise interference and inaccurate estimation of oscillatory modes. To address these challenges, this paper proposes a novel method that integrates a fractal-ridge-optimized multi-synchrosqueezing transform (MSST) with a fitting-error-minimized matrix pencil (MP). Specifically, a fractal dimension strategy based on box-counting analysis adaptively selects the number of reconstruction ridges, enabling reliable separation of signal and noise components. In addition, a quadratic correction algorithm refines ridge trajectories to suppress spectral distortions and improve time–frequency concentration. Finally, the MP algorithm is enhanced with a fitting-error minimization criterion for adaptive order determination, preventing spurious modes and ensuring accurate parameter estimation. Experimental results show that, for input SNR = 10 dB, the proposed approach improves output SNR by 10.08 dB, while the average estimation errors are as low as 0.001 for frequency and damping, and about 0.008 for amplitude and phase. These findings demonstrate that the proposed method substantially improves noise resistance and parameter accuracy, providing a reliable tool for practical LFO monitoring in wide-area power systems.