Power Quality Disturbance Classification Based on Advanced Autoencoder Networks

Power quality disturbances (PQDs) must be adequately detected and handled in the forthcoming era that includes smart grids, smart energy meters, and integration of renewable energy sources with grids. In this research, an approach for classifying PQDs and locating them without the need of sophisticated signal processing methods or sophisticated classifiers is provided. The suggested advanced autoencoder (AAE)-based network solves the issue of extracting optimal, resilient, and robust features from the PQD signal. First PQD image is convoluted with the Gabor filter which yields the Gabor features. Then, using an advanced autoencoder-based deep neural network with attention layers, important and optimum features are retrieved from these characteristics, and classification is completed using a simple SoftMax-based classifier. In this paper nine different types of PQDs are classified, encompassing single PQDs. In terms of intelligent feature selection, the proposed AAE network is similar to deep learning-based networks, but it needs far less data sets. Therefore, the PQD classification operation is completed with comparatively reduced processing complexity and time.