Complex power quality disturbances classification via curvelet transform and deep learning

This paper presents a novel approach to detect and classify the power quality disturbance (PQD) signals based on singular spectrum analysis (SSA), curvelet transform (CT) and deep convolutional neural networks (DCNNs). SSA is a non-parametric technique, does not require any supposition to generate the observed signal, and provides an effective way to recognize weak transient PQ signal. Fast discrete curvelet transform (FDCT) is an efficient method compared to wavelet transforms. Firstly, PQD signals are decomposed using SSA and FDCT methods. Initial six and three levels decomposition of the SSA and FDCT are used as features of PQD respectively. Finally, DCNNs based classifier and multiclass support vector machines (SVMs) classifier are used for classification of single and complex PQDs. For validation of the proposed algorithm, thirty-one categories of real and synthetic PQD waveforms are considered. The proposed scheme is tested and the results are recorded. The results of proposed SSA-FDCT-DCNN (SFD) based classifier are compared to the results of multiclass SVM based and other existing methods The achieved results show that the SFD classifier is more proficient than the multiclass SVM and other present methods. In addition, the proposed SFD based classifier can be efficiently used to classify the single and complex PQ disturbances.