Classification of Power Quality Disturbances Using Semi-supervised Deep Belief Networks

Modern power system has to deal with many power quality related issues due to increasing renewable integration with grid and increasing non-linear power electronics based load on the user side. Power quality disturbances (PQDs) affects the overall performance of power system technically as well as economically. Hence, classification of PQD is important and high classification accuracy is required for the same. In this work deep learning based PQ classifier model is presented. This model is based on semi supervised deep belief networks (SDBN). DBN is known for obtaining higher classification of accuracy but is majorly unsupervised learning-based model. In SDBN, supervised information is embedded into DBN learning process making the learning to be semi-supervised in a way. As compared to other popular models improved classification accuracy under noisy conditions is obtained along with less processing time for PQ classification. The addition of supervised learning layers in DBN helps extract much more relevant and strong features from PQ data.