ES-PINN Enhances Power System Stability Assessment.

Transient-stability assessment is a critical task in power system operations, determining a system's ability to recover after disturbances and preventing widespread outages. A key metric in this assessment is the Critical Clearing Time (CCT), which defines the maximum duration a fault can persist before the system loses synchronism. Estimating CCT reliably is challenging due to the complex, dynamic nature of fault-clearing processes, often requiring numerous simulations. This research introduces Event-Structured Physics-Informed Neural Networks (ES-PINN) to address these challenges. ES-PINN is designed to align its internal representation with the distinct phases of power system dynamics: pre-fault, fault-on, and post-clearing swing. A crucial feature is its enforcement of exact state chaining across these event interfaces, ensuring continuity and physical consistency. The framework defines a smooth, trajectory-induced stability margin, which provides a differentiable approximation of the CCT boundary. This differentiability allows for accurate boundary extraction, local sensitivity analysis, and even direct CCT prediction. The authors also prove a local residual-to-trajectory-to-CCT error estimate, demonstrating that exact event chaining eliminates separate state-interface defect terms. Experiments on standard IEEE bus systems show ES-PINN consistently outperforms neural-surrogate baselines in trajectory and stability-boundary accuracy, while also being computationally efficient.