Multistage convolutional neural network for power quality disturbances classification

The increasingly complex power network and the massive use of power electronics and non-linear loads in both industrial and residential areas have led to increased power quality risks. Maintaining and improving power quality (PQ) requires proper identification and classification of power quality disturbances (PQDs). In traditional methods, feature engineering is the key to the identification and classification of PQDs, but it is a strenuous, tedious, and time-consuming endeavor. This paper proposes the development of convolutional neural networks to identify and classify various types of PQDs. The proposed architecture has five stages where each stage contains a 2×1-D convolutional layer, a maxpool layer, and a batch normalization layer. This architecture can realize the extraction and selection of the best features automatically while classifying 15 types of PQDs.