An S-transform based neural pattern classifier for non-stationary signals

The paper presents a new approach for the classification of non-stationary signal patterns in an electric power network using a modified wavelet transform and neural network. The wavelet transform is phase corrected to yield a new transform known as the S-transform, which has an excellent time-frequency resolution characteristic. The phase correction absolutely references the phase of the wavelet transform to the zero time point, thus assuring that the amplitude peaks are regions of stationary phase. Once the features of a noisy time varying signal during steady state or transient conditions are extracted using the S-transform, they are passed through either a feedforward neural network or a probabilistic neural network for pattern classification. The average classification accuracy of the noisy signals due to disturbances in the power network is of the order 98%.