ML-Based Edge Application for Detection of Forced Oscillations in Power Grids

This paper presents a Machine Learning (ML) solution deployed in an Internet-of-Things (IoT) edge device for detecting forced oscillations in power grids. We base our proposal on a one-dimensional (1D) and two-dimensional (2D) Convolutional Neural Network (CNN) architecture, trained offline and deployed on an Nvidia Jetson TX2. Our work also shows the advantages of optimizing the CNNs models, after training, using TensorRT, a library for accelerating deep learning inference in real-time. Both real-world and synthetic measurement signals are employed to validate the applicability of the proposed approach.