Prediction of solitary wave forces on sea-crossing bridge decks using stacking ensemble and CGAN model

Tools

Xu, Guoji and Xu, Lele and Song, Yi and Wang, Jinsheng and Dai, Jian (2026) Prediction of solitary wave forces on sea-crossing bridge decks using stacking ensemble and CGAN model. Marine Structures, 109. 104107. ISSN 0951-8339 (https://doi.org/10.1016/j.marstruc.2026.104107)

[thumbnail of Xu-etal-MS-2026-Prediction-of-solitary-wave-forces-on-sea-crossing-bridge-decks]
Preview
 Text. Filename: Xu-etal-MS-2026-Prediction-of-solitary-wave-forces-on-sea-crossing-bridge-decks.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (11MB)| Preview

Abstract

Sea-crossing bridges are the critical infrastructure of significant economic, social, and strategic importance, as they provide vital connections across continents, islands, and straits. Accurate and efficient prediction of extreme wave loads is essential for optimizing design and ensuring operational safety under harsh marine conditions. This study develops and validates a CFD numerical model in OpenFOAM for box-girder bridge decks subjected to solitary waves, from which a systematic dataset of wave force time histories is generated across varying wave heights, water depths, and submergence coefficients. For predicting extreme wave forces, a Bayesian Hyperparameter Optimized Stacking (BHO-Stacking) ensemble model is constructed, achieving superior accuracy and robustness across multiple prediction tasks. To address full-time history prediction, one-dimensional wave force series are transformed into two-dimensional Gramian Angular Field (GAF) feature maps, serving as generation targets for a Conditional Generative Adversarial Network (CGAN). By integrating peak force predictions from the Stacking model, the generated GAF maps are temporally reconstructed into physically interpretable wave force curves. Comparative evaluations with baseline models (e.g., LSTM, BiLSTM) demonstrate that the proposed “Stacking + GAF + CGAN” framework outperforms alternatives in capturing wave force peaks, global temporal morphology, and high-frequency slamming features. The findings present a novel and effective approach for predicting wave loads on sea-crossing bridges, offering a tool for their safe design and operation.

ORCID iDs

Xu, Guoji, Xu, Lele, Song, Yi, Wang, Jinsheng ORCID logoORCID: https://orcid.org/0000-0003-1253-3050 and Dai, Jian;
CORE (COnnecting REpositories)