Electric Power Quality Disturbances Classification based on Temporal-Spectral Images and Deep Convolutional Neural Networks

We propose a deep learning based technique for power quality disturbances (PQD) detection and identification that aims at mimicking the reasoning of human field experts. This technique consists of processing small-size images containing superimposed time and frequency representations of the electric signal. The classification of PQD is performed with a convolutional neural network (CNN) trained with synthetic signals containing various single and multiple PQDs. Simulation results show that our technique is able to detect and identify with a high accuracy, in addition to pure sinusoidal, eight single PQDs and 20 of their combinations (up to four PQDs in the same signal) even in the presence of noise. Features such as lower computational load and simplicity while maintaining high performance sets the proposed technique apart from previous ones.