An intelligent risk evolution simulation method for grid operation in offshore wind power export systems

Abstract

Offshore wind power export systems couple weather-driven generation, weak offshore grids, voltage-source-converter high-voltage direct-current links, submarine cables, and onshore networks. Their operational risk can therefore evolve through interacting electrical, thermal, and control processes that are difficult to reproduce fast enough for online decision support. This paper proposes a physics-constrained temporal graph risk simulator (PC-TGRS) for intelligent risk evolution simulation. Multirate phasor, supervisory-control, converter, protection, meteorological, and distributed fiber-optic cable sensing channels are mapped to a dynamic attributed graph. A conditional scenario engine generates physically admissible wind ramps, cable derating, converter saturation or blocking, and AC/DC contingencies. A topology-masked graph attention encoder and gated temporal module then predict state trajectories, while power-balance, frequency, converter limit, and cable-thermal residuals regularize the surrogate. The resulting calibrated risk head jointly estimates risk class, continuous severity, vulnerable components, and warning lead time. On a reproducible reduced-order benchmark containing 14,400 trajectories for a 1.2-GW offshore wind cluster connected through a ±320-kV VSC-HVDC export link, PC-TGRS achieved a macro-F1 of 0.962, an AUROC of 0.989, and a risk-score mean absolute error of 0.028. It provided a median 1.84-s warning lead and evaluated 1,000 candidate scenarios in 0.41 s, approximately 150 times faster than the reference time-domain simulation. The results demonstrate an auditable route from multimodal observation to automated preventive-control prioritization.