Reliable motion prediction of Floating Offshore Wind Turbine (FOWT) platforms under stochastic metocean conditions is critical for structural safety, fatigue assessment, and operational control. Although Recurrent Neural Networks (RNNs) effectively capture temporal dynamics, their deterministic nature lacks uncertainty quantification, potentially leading to unforeseen failures, over-engineered designs, and increased maintenance costs. This study addresses this gap by introducing an Adaptive Conformalized RNN Ensembles (ACRNNE) framework that builds on conformal prediction and bootstrap-based RNN ensembles. The framework integrates three adaptive mechanisms: (1) an inverse-error weighting strategy that dynamically adjusts the contribution of each ensemble member at every prediction step, (2) a residual pool with weighted conformal inference and (3) an online update of the coverage level driven by the observed outcomes of earlier forecasts. Unlike conventional RNN models, the proposed method provides prediction intervals that adapt to the current metocean condition. The ACRNNE framework is thoroughly benchmarked against four established uncertainty-quantification techniques. ACRNNE achieves more consistent empirical coverage across the diverse metocean conditions than the evaluated baselines, with the largest gains for high, long-period waves. Furthermore, evaluation across multi-degree-of-freedom responses demonstrates robust performance with both LSTM and GRU architectures. Overall, ACRNNE enables uncertainty-aware motion prediction of FOWT platforms through empirically calibrated conformal intervals, supporting risk-informed decision-making in their operation and management.

Adaptive Conformalized RNN Ensembles for Uncertainty-Aware Motion Prediction of Floating Offshore Wind Turbines in Stochastic Marine Environments / Tamuly, P., Del Ser, J., Fathnejat, H., Nava, V.. - In: RESULTS IN ENGINEERING. - ISSN 2590-1230. - 32:(2026). [10.1016/j.rineng.2026.114034]

Adaptive Conformalized RNN Ensembles for Uncertainty-Aware Motion Prediction of Floating Offshore Wind Turbines in Stochastic Marine Environments

Vincenzo Nava
2026

Abstract

Reliable motion prediction of Floating Offshore Wind Turbine (FOWT) platforms under stochastic metocean conditions is critical for structural safety, fatigue assessment, and operational control. Although Recurrent Neural Networks (RNNs) effectively capture temporal dynamics, their deterministic nature lacks uncertainty quantification, potentially leading to unforeseen failures, over-engineered designs, and increased maintenance costs. This study addresses this gap by introducing an Adaptive Conformalized RNN Ensembles (ACRNNE) framework that builds on conformal prediction and bootstrap-based RNN ensembles. The framework integrates three adaptive mechanisms: (1) an inverse-error weighting strategy that dynamically adjusts the contribution of each ensemble member at every prediction step, (2) a residual pool with weighted conformal inference and (3) an online update of the coverage level driven by the observed outcomes of earlier forecasts. Unlike conventional RNN models, the proposed method provides prediction intervals that adapt to the current metocean condition. The ACRNNE framework is thoroughly benchmarked against four established uncertainty-quantification techniques. ACRNNE achieves more consistent empirical coverage across the diverse metocean conditions than the evaluated baselines, with the largest gains for high, long-period waves. Furthermore, evaluation across multi-degree-of-freedom responses demonstrates robust performance with both LSTM and GRU architectures. Overall, ACRNNE enables uncertainty-aware motion prediction of FOWT platforms through empirically calibrated conformal intervals, supporting risk-informed decision-making in their operation and management.
File in questo prodotto:
File Dimensione Formato  
main (2).pdf

accesso aperto

Tipologia: 2a Post-print versione editoriale / Version of Record
Licenza: Creative commons
Dimensione 4.53 MB
Formato Adobe PDF
4.53 MB Adobe PDF Visualizza/Apri
Pubblicazioni consigliate

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11583/3018006