Prediction of Hall anchor penetration depth using machine learning models

Abstract

The depth of anchor penetration into the seabed is crucial for evaluating the safety of underwater structures. This study focused on predicting the penetration depth of Hall anchors. Through physical model tests and literature collection, a comprehensive database containing 336 groups of data was established, of which 122 groups of physical test data included 5 anchor masses, 11 sediment strengths, and 12 touchdown velocities. The performance of four machine learning models and three empirical formulae was evaluated in terms of prediction accuracy, stability, and generalization ability. The RFR model demonstrated the highest reliability, achieving R² values over 95 %, with an MAE of 0.116m and an RMSE of 0.196m. The FNN and DTR models also showed high accuracy and generalization ability with R² values of 95 % and 90 %, respectively, but showed instability in random partitioning of the dataset and repeated predictions. The LR model performed poorly, while the empirical formulae had moderate performance with R² values of approximately 60 %. These findings highlight the potential of machine learning models as effective tools for predicting anchor penetration depth and assessing subsea infrastructure.