A hybrid POD-LSTM framework for efficient vortex-induced vibration prediction in marine risers via multi-fidelity data fusion

Accurate prediction of vortex-induced vibrations (VIV) in marine risers is critical for ensuring structural integrity and preventing fatigue failures in offshore engineering applications. Conventional high-fidelity numerical models suffer from prohibitive computational costs, while low-fidelity approximations often lack sufficient accuracy for reliable predictions. This study proposes a novel reduced-order modeling framework integrating Proper Orthogonal Decomposition (POD) for dimensionality reduction and Long Short-Term Memory (LSTM) networks for temporal dynamics learning. The key innovation lies in constructing a multi-fidelity surrogate that leverages sparse high-resolution data to anchor physical accuracy, while efficiently mapping abundant low-fidelity simulations through transfer learning. Validated against numerical experimental benchmarks, the hybrid model achieves an average prediction accuracy of 92.7%, as quantified by a normalized root-mean-square error (NRMSE) metric evaluated in the physical space against high-fidelity reference solutions. This approach significantly alleviates the data scarcity challenge in high-fidelity modeling and enables practical vibration forecasting for complex fluid-structure interaction systems.