Deep Learning Prediction of Settlement and Stress Fields in Geosynthetic Reinforced Pile Supported Soft Ground

A probabilistic data-driven framework for modelling clay consolidation and settlement behavior

Predicting consolidation settlement in large-scale reclamation projects remains difficult because laboratory consolidation test data are sparse in space and natural marine clays show strongly non-linear structured behavior. This study develops a probabilistic data-driven framework to reconstruct spatially continuous e–log p′ and log k–e relationships from sparse consolidation test data. A deep neural network is combined with repeated K-fold cross-validation to estimate the ensemble mean response and its 95% confidence interval. The framework was trained using consolidation test data from 49 boreholes at Kobe Airport and was evaluated using two independent blind-test boreholes that were not used in model development. The predicted mean curves reproduced the main features of the observed compression and permeability responses, including depth-dependent yield behavior and post-yield changes in compressibility. The engineering applicability of the framework was examined through settlement analysis at monitoring point KC-1, where no site-specific borehole data were available. In this analysis, soil deformation was treated as one-dimensional, whereas pore-water flow was modeled as two-dimensional. The predicted material relationships were used as input. The calculated settlement history reproduced the main observed trend, while the late-stage difference from the measurements showed the importance of deeper strata outside the present modeling scope. The proposed framework provides a practical way to interpolate consolidation behavior in space with quantified ML-related uncertainty for large reclamation projects with similar geological and data conditions.

Diaphragm-Wall Settlement Prediction and Relative Anomaly Screening for Deep Excavations Using Multi-Model Comparison and Intelligent Optimization

Deep excavations are high-risk geotechnical activities, and accurate prediction of diaphragm-wall settlement is important for construction monitoring and deformation control. This study investigates cumulative vertical settlement at 23 diaphragm-wall monitoring points from the deep excavation of Tianjin Goldin Finance 117 in Tianjin, China. Seven prediction models—a naïve persistence model, autoregressive integrated moving average (ARIMA), K-nearest neighbors (KNN), multilayer perceptron (MLP), gated recurrent unit (GRU), Transformer, and XGBoost—were evaluated using a unified five-fold rolling-origin expanding-window validation scheme. GRU achieved the best overall baseline performance, with a mean R2 of 0.9172, a mean absolute error (MAE) of 0.1009 mm, a root mean square error (RMSE) of 0.1341 mm, and a mean absolute percentage error (MAPE) of 0.5875%. GRU was subsequently optimized using the crow search algorithm (CSA), the genetic algorithm (GA), and the whale optimization algorithm (WOA). GRU-WOA achieved the best numerical performance, with a mean R2 of 0.9289 and an RMSE of 0.1215 mm. Relative anomaly levels were further identified from predicted settlement-change rates to characterize temporal concentration and spatial clustering of settlement-change activity. The proposed framework can support priority inspection and targeted monitoring, although the resulting anomaly levels represent project-relative statistical deviations rather than code-based engineering risk classes.

Bayesian-Optimized Ensemble Machine Learning for Predicting Settlement of Cohesionless Soil Under Loads

A FIELD-CALIBRATED NUMERICAL–EMPIRICAL FRAMEWORK FOR DEEP EXCAVATION DESIGN IN SOFT ALLUVIAL SOILS OF İZMIR, TÜRKIYE

In densely populated urban environments, deep excavations in soft ground present critical geotechnical challenges due to high groundwater table and weak cohesive soils. Diaphragm walls supported by struts or tieback anchors are frequently used to stabilize such excavations, yet predicting deformation behavior remains complex and highly site-dependent. This study evaluates the performance of four deep excavation projects in İzmir, Türkiye, where excavation depths reached up to 17 meters within soft to medium alluvial deposits.A performance-based back-analysis methodology is implemented, where numerical models are calibrated using field-monitored inclinometer data. A non-linear soil constitutive model with stress-dependent stiffness and hardening behavior is employed, and the elastic stiffness modulus (E₅₀) is iteratively adjusted to reflect actual wall displacements. A site-specific empirical correlation is developed between E₅₀ and SPT-N₆₀, resulting in a power-law model with strong predictive capability for local soil conditions.Findings demonstrate that conventional empirical stiffness correlations often fail to capture the deformation response of deep excavations in İzmir's geologic setting. The calibrated models offer improved insight into wall-soil interaction and enhance the reliability of deformation predictions. The study promotes a practical framework integrating numerical modeling with field monitoring for safer, performance-based design in soft urban ground.

Machine Learning Prediction of the Ground Reaction Curve in Sand with the MATLAB GUI Platform

The proposed framework combined a curated database, neural network-based curve prediction, and hyperparameter optimization, providing a robust approach for evaluating the soil arching effect, providing a robust approach for evaluating the soil arching effect.

Field-Constrained Dual-Correction Model for Predicting Casing Stress During Multi-Stage Hydraulic Fracturing in Deep Coalbed Methane Horizontal Wells

Deep coalbed methane reservoirs exhibit strong heterogeneity and complex stress environments, making accurate prediction of casing loads during multi-stage hydraulic fracturing challenging. This study developed a coupled prediction framework integrating three-dimensional geomechanical modeling, modified induced stress calculation, and numerical simulation. The geomechanical model was constructed using logging, drilling, rock mechanics, and in situ stress data and validated against fracture monitoring results. The simulated fracture half-length and stimulated area differed from the monitored values by less than 10% and 8%, respectively, with a spatial matching degree exceeding 92%. A modified analytical model was then established by introducing correction coefficients for fracture net pressure and stress propagation. Among the results obtained using five parameter inversion methods, the sparrow search algorithm achieved the highest fitting accuracy, with an (R2) of 0.9371 and an RMSE of 0.3761, yielding (A = 0.7330) and (B = 0.9238). These coefficients indicate an approximately 26.7% reduction in effective net pressure and enhanced attenuation of induced stress in heterogeneous, cleat-developed coal seams. Furthermore, a multi-parameter casing stress model was developed by coupling treatment scale, injection rate, fracture spacing, and stage number. Sensitivity analysis showed that the number of fracturing stages and injection rate were the dominant factors, followed by fracture spacing and treatment scale. The proposed framework quantitatively characterizes casing stress evolution and facilitates casing load assessment under different multi-stage fracturing conditions.