Recognition of voltage sag causes using fractionally delayed biorthogonal wavelet

In this work, a new fractionally delayed biorthogonal wavelet is designed for recognition of voltage sag causes. This work addresses the classification of voltage sag causes into three categories, i.e., fault events (namely line-to-ground fault, line-to-line fault, double-line-to-ground fault and symmetrical fault), induction motor starting and transformer energization. Fractionally delayed biorthogonal Coiflet wavelet of order 2 (named as Coiflet fraclet) is designed using Lagrange interpolation and employed for the multiresolution analysis of voltage sag signals. To achieve higher classification accuracy, event information is derived from both time-frequency domain and frequency domain. Frequency information is obtained by applying multiple signal classification (MUSIC) algorithm on voltage sag signals. Both the time-frequency domain and frequency domain features are concatenated to implement the feature level fusion for strengthening the feature set. The feature set obtained after feature level fusion is used for training the recursive reduced kernel based extreme learning machine which addresses the problem of highly complex structure due to huge dataset of large dimensions. The classifier has shown robustness to the presence of noise also and classified the voltage sag causes with high accuracy.