A particle swarm optimization algorithm based demodulation method for amplitude frequency modulation characteristics of power system oscillation signals is proposed to address the problem of power system oscillation caused by the continuous increase in the proportion of new energy grid connected. Firstly, starting from the physical mechanisms of formation and evolution of AC electrical quantities, the constraint relationships of oscillation signals under disturbances are clarified, revealing the modulation patterns of their amplitudes and frequencies over time. Unlike traditional harmonic superposition methods based on Fourier decomposition or wavelet analysis, this study emphasizes that oscillation signals are not formed by static harmonic superposition but rather result from a continuous modulation process shaped by the system’s dynamic closed-loop action, which carries greater physical significance. On this basis, a PSO-based amplitude⁃frequency modulation feature extraction method is designed. By minimizing the deviation between the fitted waveform and the actual waveform, the amplitude⁃frequency modulation parameters that determine the oscillation process are extracted. Simulation results show that the proposed method can accurately reflect the oscillation patterns of typical converter-interfaced power systems, with the fitted waveforms highly consistent with the original signals. Dynamic model experiments further validate the applicability and effectiveness of the proposed algorithm in practical systems.