Critical Clearing Time Computation in Power System Transient Stability Analysis: a Physics-Informed Neural Network approach

This paper presents an original application of physics-informed neural networks (PINN) to the problem of computing the critical clearing time in power system rotor-angle transient stability studies. We first analyze the accuracy of a properly trained PINN in terms of its ability to correctly reproduce the dynamics of a generator-infinite bus system evolving from initial conditions next to the boundary of the region of attraction of its unique stable equilibrium. We then integrate the PINN in a mixed-integer linear programming problem that, assuming no prior knowledge about the region of attraction, provides the critical clearing time of the system and, as a by-product, the system trajectories characterizing the full fault and clearing process. The simulation results are presented to validate the proposed approach.