Power signal disturbance identification and classification using a modified frequency slice wavelet transform

This study presents a novel approach to localise, detect and classify non-stationary power signal disturbances using a modified frequency slice wavelet transform (MFSWT). MFSWT is an extension of frequency slice wavelet transform (FSWT), which provides frequency-dependant resolution with additional window parameters for better localisation of the spectral characteristics. An advantage of the MFSWT is attributed to the fact that the modulating sinusoids are fixed with respect to the time axis, whereas a localising scalable modified Gaussian window dilates and translates. Several practical power signals are considered for visual analysis using MFSWT, and the disturbance patterns are appropriately localised with unique signature corresponding to each type. This work also evaluates the detection capability of the proposed methodology and a comparison with earlier FSWT and Hilbert transform to show the superiority of proposed technique in detecting the power quality disturbances. A probabilistic neural network (PNN) based classifier is used for identifying the various disturbance classes. The spread parameter of the Gaussian activation function in PNN is tuned and its effect on the classification at different strengths of noise is studied.