Deepwater oil and gas development imposes stringent safety requirements, exemplified by the “Haiji” jacket platforms operating at 200–400 m water depths. This study proposes a machine learning–based digital twin framework for axial force prediction in deepwater jackets. The framework integrates multi-source monitoring data with structural response features. A LightGBM model is first trained using simulation data to obtain the optimal predictive performance. The model is then refined with measured data through parameter optimization and feature construction, thereby improving generalization and engineering applicability. Results show that the method effectively captures nonlinear relationships between environmental parameters and structural responses, achieving about 10% prediction error on measured data. The approach significantly enhances predictive accuracy and computational efficiency, providing a practical and intelligent solution for lifecycle safety management of deepwater jacket platforms.
Zhou et al. (Sun,) studied this question.