GENERATION OF SYNTHETIC EXAMPLES USING GENERATIVE ADVERSARIAL NETWORKS (GAN) TO EXTEND A DATABASE OF FAULT SIGNALS ON POWER DISTRIBUTION LINES

The detection and classification of the type of fault is an essential technique for the improvement of electricity grids due to its potential to improve the reliability of supply and, therefore, its quality. This paper reports a method to obtain an extended database of fault signals in order to use Neural Networks (NN) to process them. The need of a large database for the training process is an inherent need for the right working of a NN. In this type of chaotic nature signals, it is impossible to record enough real ones and, even simulating is near unfeasible task due to the variety of the causes that produces faults events. The proposed solution is to obtain a short database of simulated signals from a real modelled electrical grid and extend this database by means of GAN. This technique simplifies the process to obtain the database of fault signals.