Scalable Physics-Informed Neural Networks for Accelerating Electromagnetic Transient Stability Assessment

This paper puts forward a framework to accelerate Electromagnetic Transient (EMT) simulations by replacing individual components with trained Physics-Informed Neural Networks (PINNs). EMT simulations are considered the cornerstone of transient stability assessment of power systems with high shares of Inverter-Based Resources (IBRs), and, although accurate, they are notorious for their slow simulation speed. Taking a deeper dive into the EMT simulation algorithms, this paper identifies the most computationally expensive components of the simulation and replaces them with fast and accurate PINNs. The proposed novel PINN formulation enables a modular and scalable integration into the simulation algorithm. Using a type-4 wind turbine EMT model, we demonstrate a 4--6x simulation speedup by capturing the Phase-Locked Loop (PLL) with a PINN. We validate all our results with PSCAD software.