Quasi-Periodic Harmonic Feature Extraction of Power Signals via Improved Scaling-Basis Chirplet Transform

The increasing randomness of source–load fluctuations and the rapid proliferation of power-electronic devices have introduced wide dynamic ranges, fast time-varying behaviors, and strong stochastic characteristics into power signals, leading to severe measurement deviations in existing metering equipment. Conventional modeling and feature analysis methods based on fixed-frequency steady-state assumptions are inadequate for characterizing such non-stationary behaviors, making the underlying causes of metering deviations difficult to identify. To address this issue, we propose a modeling and dynamic time–frequency feature extraction method for complex non-stationary power signals. First, the operating characteristics of power equipment are analyzed to identify the fundamental non-stationary features of power signals, based on which a quasi-periodic harmonic signal model is established. Then, the scaling-basis chirplet transform is employed to intuitively represent the time–frequency structure, while a ridge detection algorithm is incorporated to quantitatively characterize the time–frequency trajectories and instantaneous amplitude features. Finally, to cope with the limited availability of power signal measurements, a non-stationary component reconstruction method based on cross-correntropy is developed. Experimental results from multiple datasets, including field-measured signals, demonstrate that the proposed method enables effective dynamic monitoring and reconstruction of non-stationary components, offering significant advantages in both time–frequency analysis capabilities and reconstruction accuracy.