A hybrid VMD-IGA-LSTM model for dynamic response prediction of jacket offshore platform

This study proposes a hybrid network for predicting the structural response of a jacket offshore platform under wave and earthquake excitations. The framework integrates three techniques: variational mode decomposition (VMD), improved genetic algorithm (IGA), and long short-term memory (LSTM) networks, which enhance the stability and precision of structural response prediction through efficient data preprocessing and optimization strategy . First, a finite element model of jacket offshore platform considering soil-structure interaction is established, and a dataset for model evaluation is constructed based on this model. Next, a single-input model is designed, and comparative experiments are conducted to validate the multi-input model's ability to capture the coupling effects between different loads. Finally, ablation experiments are carried out to investigate the impacts of VMD and IGA on model performance. The results demonstrate that the VMD-IGA-LSTM model excels in prediction accuracy, training efficiency and robustness, showing strong prediction capability and stability for the dynamic responses of jacket offshore platform subjected to the combined seismic and wave loads. • A deep learning-based structural response prediction model for jacket offshore platform is proposed, integrating variational mode decomposition, improved genetic algorithm and LSTM networks. • A finite element model of jacket offshore platform including soil-structure interaction effects is established, and a database is created to evaluate the prediction model's performance. • The effectiveness of the model in handling the coupling effects of wave and seismic loads is validated through comparative experiments with a single-input model.