Long-term water entry impact force forecasting using LSTM-RF with recursive prediction strategy

作者：Xinyu Wu, Xiaoyi Li, Chang Yuan, Donghai Zeng, Jianxing Zhang, Baoren Li, Gang Yang · 发表于：Ships and Offshore Structures · 年份：2025 · DOI：10.1080/17445302.2025.2517375 · 被引用次数：3 · 研究领域：Tropical and Extratropical Cyclones Research、Fluid Dynamics Simulations and Interactions、Hydrological Forecasting Using AI

Transient impact forces during water entry critically affect structural safety and entry motion stability. This study proposes an LSTM-RF fusion model for rapid prediction of initial-stage water impact loads. Leveraging numerical simulation data, an LSTM network with recursive strategy predicts long-term impact loads, while a Random Forest algorithm compensates for prediction errors, enhancing accuracy and generalization. Results demonstrate significant improvements over conventional LSTM: Mean Absolute Error, Root Mean Square Error, and Relative RMSE decrease by 60%, 54%, and 44% respectively, with peak prediction errors consistently below 10.5%. Computational efficiency increases dramatically – execution time reduces from 8 hr 42 min to 2 min 34 s. The model maintains predictive accuracy while substantially reducing computational costs.