Abstract: Perception and prediction of soft-ground consolidation behavior constitute a prerequisite for stability analysis and settlement calculation of soft foundations; accurate evaluation of the degree of consolidation hinges on the reliable determination of consolidation parameters and drainage boundary conditions. To identify consolidation parameters and consolidation characteristics of soft soils, a physics-informed neural network (PINN) is developed for the soft-foundation consolidation equation. The model uses the automatic differentiation capability of neural networks to represent the governing differential equation. Continuous-drainage boundary conditions are introduced to account for time-dependent pore-water pressure at the boundaries. A composite loss function integrating the physical equation, continuous-drainage boundaries, and initial conditions is constructed, while pore-water pressure dissipation data further incorporated as a data-driven term. Consequently, a physics- and data-driven model is established for the integrated analysis and prediction of soft-ground consolidation behavior. Through model training and hyper-parameter optimization, in-situ consolidation coefficients of soft soils and parameters of continuous-drainage boundaries are inversely identified, enabling the digital-intelligent prediction of consolidation behavior. After verifying the inversion and prediction accuracy of the model, the influences of piezometer layout, monitoring period, and acquisition frequency on parameter-identification accuracy are investigated. Results indicate that placing piezometers in locations with pronounced pore-water pressure variations markedly improves identification accuracy. Parameter identification accuracy also increases with longer monitoring periods and higher data acquisition frequencies. By effectively integrating physical mechanisms, the proposed model accurately identifies multiple unknown consolidation parameters using only short-term pore-pressure observations. It also provides integrated predictions of pore-pressure dissipation and consolidation development. This study offers a reliable and efficient approach for soft-ground treatment design, construction control, and post-construction settlement prediction.

Key words: continuous drainage boundary, physics-informed neural network, consolidation parameters inversion, consolidation degree deduction, data-mechanism integration