Wall deflection prediction for braced excavations in soft clay overlying stiff soil

Abstract Urbanization has intensified the demand for deep excavations, particularly in soft clay environments, where controlling wall deflection poses significant engineering challenges due to the low strength, high compressibility, and sensitivity of the soft clay. Traditional methods relying on empirical and mechanical calculations often fall short under such complex geological conditions, necessitating advanced predictive techniques. This study investigates wall deflection during braced excavations in soft clay overlying stiff soil using a combination of finite element analysis (FEA) and machine learning approaches. A Hardening Soil model was employed to simulate the soil, incorporating six critical variables, including the excavation width, soft clay thickness, soil undrained shear strength, soil modulus ratio, diaphragm wall width, and excavation depth. A total of 768 cases were analyzed, generating a comprehensive deformation database. These data were further supplemented with field monitoring records, creating a robust dataset of 906 samples for machine learning analysis. Four machine learning algorithms, Decision Tree, Random Forest (RF), Polynomial Regression, and XGBoost, were evaluated for their predictive accuracy using metrics such as Root Mean Square Error, Coefficient of Determination, and Mean Absolute Error. Results highlight the RF model’s superior performance, with stable accuracy across training and test datasets, while XGBoost showed promising results with slightly reduced generalization. Spearman correlation analysis revealed strong correlations between deflection and variables like soft clay thickness and soil undrained shear strength. A series of design charts has been developed and verified. This study not only demonstrates the efficacy of integrating FEA and big data techniques but also identifies key factors influencing wall deflection. These findings provide insights for advanced predictive methodologies in geotechnical engineering.