COMPRESSIVE DEFORMATION CHARACTERISTICS AND PREDICTION MODEL OF SULFATE SALINE SOIL

Saline soil exhibits distinctive deformation characteristics due to the soluble salts within, and its compressive deformation mechanism and predictive model are of crucial significance for engineering construction in cold-arid regions. This study focuses on sodium sulfate saline soil from the Datong Basin, systematically analyzing the effects of initial moisture content (16%, 20%, and 24%), salt content (0.3%~20%), and overburden pressure (12.5~1600 kPa) on compressive deformation through uniaxial compression tests. Machine learning prediction models were developed using Random Forest (RF), Gradient Boosted Decision Tree (GBDT), Support Vector Machine (SVM), and Back Propagation Neural Network (BPNN). Key findings include: (1) Compressive deformation increases linearly with initial moisture content (up to an 80% increment), while salt content exhibits a nonlinear influence: deformation increases by 30% as dissolved salts expand pore fluid volume before solution saturation, then decreases by 15% after saturation due to the formation of a crystalline skeleton, with peak deformation occurring at the saturation threshold. (2) The void ratio first decreases then increases with salt content under constant pressure, showing the maximum compressibility coefficient at the saturation state. Soils with 24% moisture and 12%~16% salt content exhibit high compressibility, while others show medium compressibility. (3) Machine learning performance comparison reveals superior predictive accuracy for BPNN, RF, and GBDT (R²≥0.97) over SVM (R²=0.904). Among them, the BPNN model demonstrates the best performance (R²=0.982, RMSE=0.145, MAE=0.099). The multi-layer network structure of the BPNN model can effectively capture the salt crystallization effect and nonlinear interactions during compression. This study advances the quantitative understanding of saline soil compression mechanisms, providing a theoretical foundation and computational support for disaster prevention in cold-arid regions and for modeling saline soil deformation.