A CPTU‐Based Prediction Method for Undrained Shear Strength With Quantified Uncertainty

The engineering properties of seabed soils are crucial to the foundation design of offshore structures. Among them, the undrained shear strength ( s u ) is a key parameter for evaluating the shear failure resistance of saturated cohesive soils. Although empirical equations based on piezocone penetration tests (CPTU) offer a convenient approach for estimating s u , their effectiveness is often limited due to the inability to capture complex data patterns. In recent years, machine learning has been widely applied in geotechnical engineering, primarily through supervised learning models. However, given the limited availability of labeled CPTU data, effectively utilizing a large amount of unlabeled data has become a critical research challenge. This study proposes a CPTU‐based s u prediction method incorporating uncertainty quantification, utilizing a self‐training semi‐supervised learning (SSL) strategy. The model is trained on 213 labeled samples and leverages 700 high‐confidence pseudo‐labeled samples selected from an extensive collection of unlabeled CPTU records. Compared to baseline models, the proposed approach reduces the MAE from 10.92 to 9.70 kPa and lowers the RMSE from 17.81 to 15.69 kPa. Uncertainty quantification enhances the stability and reliability of the predictions. SHapley Additive exPlanations (SHAP)‐based interpretability analysis reveals that tip resistance ( q c ) is the most influential input in predicting s u . The results provide a reliable tool for the design and optimization of offshore wind turbine foundations.