Data-Driven Prediction and Mechanism Analysis of Riser Vibration Response Under Multi-Parameter Coupling

To investigate the dynamic response of marine risers under multi-parameter coupling effects, physical model tests of marine risers are designed and performed in this study. The tested parameters consist of structural parameters (materials, boundary condition and top tension) and environmental flow parameters (flow velocity, wave height and wave period). Three riser specimens including aluminum tube (Al), polymethyl methacrylate tube (PMMA) and unplasticized polyvinyl chloride tube (UPVC) are adopted to qualitatively simulate metallic risers and composite risers with distinct stiffness differences. A total of 216 sets of vibration test data are acquired from the indoor flume experiments. Based on experimental data, the individual effects of various parameters on the vibration amplitude and dominant frequency of risers are first analyzed via single-factor comparative tests. Subsequently, an XGBoost machine learning model is established to predict the cross-flow vibration amplitude of risers; stratified sampling and 5 × 5-fold cross-validation are utilized to optimize model hyperparameters. The model achieved a high goodness-of-fit of R2 = 0.9296 during the development phase (training and internal validation), demonstrating favorable prediction accuracy. Finally, the SHAP interpretability method is introduced to quantify the contribution weights of individual parameters as well as the interaction mechanisms under multi-parameter coupling. Both single-parameter and multi-parameter analyses are performed, with representative results such as material = −3.965 and flow velocity = 2.085. The results indicate that the riser vibration amplitude exhibits a prominent negative correlation with material properties, boundary conditions, and top tension, while showing a positive correlation with flow velocity, wave height, and wave period.