PQ Disturbances Dataset

In order to prepare the PQ disturbances dataset, for PQ disturbances classification projects using deep learning, we consider a total of 12 disturbances, i.e., sag, swell, interruption, flicker, harmonics, transients, Swell with harmonics, Sag with harmonics, interrupt with harmonics, flicker with harmonics, swell with flicker; sag with flicker. All of these signals are created in MATLAB using a variety of parameters. They are then broken down into detail [d1, d2, d3, d4, d5, d6, d7, d8] and approximate coefficients (A8) using the Daubechies mother wavelet at level 8. The complete PQ dataset consists total 750 samples and each sample has 72 features. Each decomposed signal yields eight features: mean, standard deviation, RMS value, energy, entropy, skewness, kurtosis, and ranges. Because there are 9 decomposed signals, the total number of features is 9 x 8 = 72.

The dimensionality of the dataset is reduced by compressing 72 input features into 64 features using an autoencoder in order to help machine learning engineers build an effective AI model, These 64 features are extracted from the latent space of the autoencoder and further reduced to 21 based on statistical analysis.