Parametric methods for time-frequency analysis of electric signals

The author presents a new approach to spectral analysis of electric signals and related problems encountered in power systems. This approach includes the use of high- resolution subspace spectrum estimation methods (such as MUSIC and ESPRIT) as replacement of widely used Fourier Transform-based techniques. The author proves that such approach can offer substantial advantages in parameter estimation accuracy, classification accuracy and many other aspects of power system analysis, especially when analyzing non- stationary waveforms. The problems treated in this work include theoretical analysis of the limitations of FFT-based analysis, problems in applications of Short Time Fourier Transform, description and characteristic properties of subspace frequency estimation methods - MUSIC and ESPRIT; estimation of the model order, theoretical development of time-varying spectrum, application of filter banks and advantages when applying to line spectra analysis, space-phasor for analysis of three-phase signals, power quality assessment using indices with practical application to waveforms from an arc furnace power supply, numerical analysis of performance of investigated methods and a novel approach to classification of power system events based on time-frequency representation and selection of "areas of interest" in time-frequency plane. The author concludes that the use of high-resolution methods significantly improves the accuracy of many parameter estimation techniques applied to power system analysis