Physics–Data Fusion-Driven Frequency Response Parameter Identification and Emergency Load Shedding in Renewable-Rich Power Systems

As renewable penetration increases, the inertia, damping, and primary frequency regulation characteristics of renewable-rich power systems become strongly time-varying, making fixed offline frequency response models increasingly difficult to maintain for emergency-control calculations. This paper proposes a physics–data fusion-driven framework for frequency response parameter identification, model validation, and emergency load shedding based on post-disturbance multi-source measurements. First, a quality-aware dual-stage LSTM-PINN fuses multi-source measurements to provide event-specific initial estimates of the disturbance magnitude and physical system frequency response (SFR) parameters. A bounded event-level local constrained refinement then aligns the dynamic parameters with the measured frequency trajectory, while differentiable SFR constraints, key response losses, and identifiability regularization improve physical consistency and parameter distinguishability. Second, local identifiability, physical parameter plausibility, and trajectory consistency are jointly evaluated to characterize the credibility of the identified SFR model. Finally, the measurement-updated controlled-SFR model determines the minimum emergency load-shedding amount satisfying the frequency nadir constraint and allocates the action to candidate buses according to electrical distance and available controllable capacity. Case studies on a modified New England 39-bus system and the renewable-rich CSEE-FS benchmark evaluate parameter identification accuracy, model credibility, robustness, generalization, control security, spatial allocation, and computational efficiency.