Empirical wavelet transform and dual feed-forward neural network for classification of power quality disturbances

This paper proposes a novel approach for classification of single and combined power quality (PQ) disturbances. The EWT-based adaptive filtering technique is employed first to decompose the signal into its individual frequency components by estimation of frequencies. The frequency estimation in this paper is done using a divide-to-conquer principle-based FFT technique and followed by an adaptive filter design. Then, some unique potential features reflecting the characteristics of disturbances are extracted from the mono-frequency components as well as the signal. A single classifier used for the classification of combined disturbances, whose characteristics are alike, gives less classification accuracy. Therefore, the use of a dual FFNN is proposed for the classification of single and combined PQ disturbances to effectively reduce the misclassification and improve the accuracy. The effectiveness of the proposed approach is evaluated on a broad range of timevarying power signals with varying degree of irregularities, noise, and fundamental frequency deviation. The results obtained for both the simulated as well as the real disturbance signals elucidate the efficiency and robustness of the proposed approach for classification of the most frequent disturbances.