Continuous-time echo state networks for predicting power system dynamics

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