Continuous-time echo state networks for predicting power system dynamics

• Published in: Electric power systems research Vol. 212; no. C; p. 108562
• Main Authors: Roberts, Ciaran, Lara, José Daniel, Henriquez-Auba, Rodrigo, Bossart, Matthew, Anantharaman, Ranjan, Rackauckas, Chris, Hodge, Bri-Mathias, Callaway, Duncan S.
• Format: Journal Article
• Language: English
• Published: Switzerland Elsevier B.V 01.11.2022; Elsevier
• Subjects: Data-driven modeling techniques; Electro-magnetic transients; Machine learning; Power system dynamics; Data-driven modeling techniques; Electro-magnetic transients; Power system dynamics; Machine learning
• ISSN: 0378-7796; 1873-2046
• DOI: 10.1016/j.epsr.2022.108562

With the growing penetration of converter-interfaced generation in power systems, the dynamical behavior of these systems is rapidly evolving. One of the challenges with converter-interfaced generation is the increased number of equations, as well as the required numerical timestep, involved in simulating these systems. Within this work, we explore the use of continuous-time echo state networks as a means to cheaply, and accurately, predict the dynamic response of power systems subject to a disturbance for varying system parameters. We show an application for predicting frequency dynamics following a loss of generation for varying penetrations of grid-following and grid-forming converters. We demonstrate that, after training on 20 solutions of the full-order system, we achieve a median nadir prediction error of 0.17 mHz with 95% of all nadir prediction errors within ±4 mHz. We conclude with some discussion on how this approach can be used for parameter sensitivity analysis and within optimization algorithms to rapidly predict the dynamical behavior of the system. •CTESNs as an approach to accelerate parameter-sensitivity time domain simulation.•An empirical examination of the accuracy of CTESN predictions.•Use of CTESNs to predict frequency dynamics following a large loss of generation.•Achieve 95% of all nadir prediction within +/- 4 mHz across the parameter space.