Machine learning approach for predicting plug heave for suction caissons in cohesionless soil

1. Home
2. Publications
3. Online Library
4. Machine learning approach for predicting plug heave for suction caissons in cohesionless soil

ABSTRACT: Suction caisson foundations are a promising technology for offshore wind turbine structures, but accurate prediction of plug heave during installation remains a critical challenge. This study presents a robust Random Forest Regression (RFR) model for heave prediction in cohesionless soils, incorporating the relative density of the soil and key parameters recorded during actual installations, such as penetration depth, penetration rate and differential pressure. The model aims to ensure plug heave remains within acceptable limits, avoiding installation refusals and maintaining appropriate conditions for subsequent grouting operations. The coefficient of determination (R²) value of the regression model illustrates the proportion of the variance in the dependent variable (heave) which is explained by the independent variables. The model defined in this paper demonstrates high accuracy and strong generalisation capabilities, achieving an R² score of 94.3% on the training set, with consistent performance in cross-validation (R² of 83.3%) and test sets (R² of 83.0%). This reliability across datasets, combined with visual analysis showing close alignment between predicted and measured heave values, demonstrates the applicability of the proposed model for plug heave predictions. Explaining approximately 83% of the variance in the test set, the model provides a solid foundation for practical heave prediction for suction caisson installations in cohesionless soil, offering potential improvements over existing empirical methods.

5th International Symposium on Frontiers in Offshore Geotechnics (ISFOG2025)

5 - Data Analytics and Machine Learning