Proceedings of the PowerUp Conference

The evolving landscape of power systems is leading to larger and more complex systems. Future power systems comprise more diverse components, exhibit faster dynamics, and are, in general, less predictable. As a result, time-domain simulations are becoming increasingly necessary for both offline studies and real-time stability assessment. This evolution renders dynamic simulations computationally more demanding, calling for improved simulation speed. Although classical solvers are reliable, their computational performance is limited by the need for iterative solution schemes. We present a modular method that allows Neural Network (NN) based surrogates of any power system component to be embedded within the decoupled Newton iterations of a time-domain solver. A case study integrates Physics-Informed Neural Networks (PINNs) into a solver and demonstrates the general applicability of the proposed approach. The results reveal specific challenges regarding PINN accuracy but highlight the promise of the framework and motivate further exploration using a wider variety of surrogate models.