Fourier Synchro squeezed-LSTM-CNN for Categorization of Power Quality Disturbances

Integration of renewable energy has resulted in an increase in the power quality disturbances (PQDs) in a power network, thus, necessitating accurate PQD categorization for guaranteed reliability and stability of the power system. In this work, the voltage signal characteristics are used for PQD event categorization using deep learning techniques. A novel architectural deep neural network is built on an LSTM-based convolutional neural network to efficiently classify various power quality disturbance signals using a support vector machine classifier. Fourier Synchro squeezed Transform (FSST) is utilized to transform the PQD signal information into 2-dimensional data which is further represented as an image. Automated feature extraction from the FSST-transformed image of the PQD signal was adopted in this work for better classification of PQDs. Using this special combination, 99.99% accuracy, 99.65% precision, and 99.63% sensitivity were achieved. The method was also validated using real-life PQD signals.