Work

Causal reliability-gated GRU sensor reconstruction for a reduced-order floating offshore wind turbine under false-data injection

Year: 2026
Type: article
Abstract: Sensor reconstruction under false-data injection (FDI) is investigated on a control-oriented, reduced-order nonlinear floating offshore wind turbine inspired by the NREL 5-MW rated scale. The model combines empirical aerodynamic maps, surge–pitch platform dynamics, a single-mass drivetrain and constrained actuators. A gated recurrent unit (GRU) predicts six sensor outputs causally from trusted environmental inputs and previously applied commands. Distinct trajectories are used for training, validation, detector calibration and testing. Offline prediction-confidence weights blend received measurements and predictions after channel-wise residual detection; global detection and full-replacement variants are compared. Across 27 held-out attack/noise scenarios, channel-gated compensation reduces the dimensionless sensor-error metric by 19.02% with a mean paired rotor-speed RMSE reduction of -0.15% relative to the unprotected benchmark. Scenario reductions range from -4.04% to +6.46%, with improvement in 11 of 27 cases; near-zero mean change does not imply scenario-wise performance preservation. Channel improvements range from -3.28% for power to 78.65% for generator torque in the base condition. Reliability gating yields lower rotor-speed RMSE than full replacement in 22 of 27 matched comparisons, but superiority over fixed partial replacement or independent predictors is not established. Single-channel tests reveal limited rotor-speed attack detection: recall is 4.42% for a rotor-speed-only attack, and the clean rotor-speed false-alarm rate is 12.48%. Reliable feedback-channel protection and guaranteed post-attack recovery are not established. Vector error identities, a conditional mean-square benchmark, controller-command sensitivity, and a finite-horizon perturbation bound explain the relationship between reconstruction and regulation. The study establishes a reproducible evaluation framework for reconstruction and controller-aware assessment; its internal ablations do not separately establish advantages of the GRU architecture and the confidence-weighting rule.
Language: English
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7232204807
FWCI:
Cites: 5
Cited by:
Related to: 10