Low-frequency oscillations (LFOs) in traction power supply systems (TPSSs) frequently arise when multiple electric trains simultaneously raise their pantographs under the same power supply arm. This phenomenon is characterized by low-frequency fluctuations (2–8 Hz) in the envelope waveforms of traction network voltage and current. It can lead to operational issues such as insufficient current acquisition and difficulties in depot entry or exit

thereby adversely affecting railway operations. Given the time-varying nature of LFOs

timely and accurate detection is critical for implementing effective mitigation strategies

with faster detection enabling improved outcomes. This paper proposes a real-time detection algorithm for LFOs in AC traction networks that integrates signal preprocessing

spectral analysis

and parameter optimization. First

the voltage signal is processed using a low-pass filter to suppress high-frequency noise. Then

a fast Fourier transform (FFT)-based spectral estimation method is applied to extract frequency-domain features. Oscillation parameter identification is triggered when the identified signal amplitude exceeds predefined thresholds in the 42–48 Hz and 52–58 Hz bands. Subsequently

during the identification stage

an LFO atom dictionary is constructed based on the FFT pre-analysis results. Finally

the matching pursuit algorithm is employed to achieve fast and accurate extraction of LFO parameters. The proposed method is validated using both simulated and real-world measurement data. Experimental results confirm its effectiveness in detecting LFOs under noisy conditions

demonstrating high accuracy and computational efficiency. The approach provides valuable insights for the threshold selection of protection devices

thereby enhancing the stability and reliability of TPSSs.