Soil-water characteristics and prediction model of GMZ bentonite under the coupled effects of multiple factors

Using the steam equilibrium method of saturated salt solutions, this study investigated the evolution of the soil-water characteristic curve (SWCC) for Gaomiaozi (GMZ) bentonite under multifactorial conditions (dry density, temperature, and salinity). Results revealed that water retention capacity positively correlated with dry density. Elevated temperatures reduced water retention by shifting SWCC downward, while increased salinity enhanced water retention capacity, particularly under low suction. Dry density exerted greater regulatory influence on SWCC than temperature, whereas salinity surpassed dry density in impact intensity. Based on the vG model, by introducing different influencing factors, a mathematical model of the SWCC that includes the coupling term of dry density, temperature and salinity was constructed. Further, based on Bayesian regularization neural networks, two different forms of SWCC prediction models have been established through machine learning training on a large amount of SWCC test data influenced by multiple factors. One is a data-driven model that takes matric suction as an input variable to predict the volumetric water content under different suction conditions. The other is a data-physics fusion-driven model that uses the proposed SWCC model as a physical constraint to predict model parameters. Both models enabled accurate SWCC prediction in complex environments using basic inputs (physical properties and environmental parameters). Experimental validation confirmed their effectiveness, demonstrating reliable predictive performance for SWCC behavior influenced by multiple factors.