Abstract: Plain earth dams are frequently constructed on soft soil foundations, and their consolidation process is crucial to engineering safety. Traditional finite element methods have limitations such as high computational overhead and mesh dependency when dealing with complex geometric boundaries and heterogeneous soil layers. This paper proposes an intelligent prediction method for pore water pressure evolution during the consolidation process of plain earth dams based on physics-informed neural networks (PINN). By embedding Terzaghi’s two-dimensional consolidation equation into the neural network architecture and combining automatic differentiation technology, a physics-driven simulation of PWP evolution is achieved. Aiming at the complex geometric boundaries of the Erwangzhuang Reservoir capacity expansion project, strategies such as boundary extraction and classification, masked stratified sampling, and a two-stage joint training approach are proposed, which significantly improve the model’s adaptability to dynamic computational domains and its training stability. The results show that, even with only sparse monitoring data, the PINN model can still accurately simulate the spatial-temporal evolution process of pore water pressure. The coefficient of determination (R2) is greater than 0.83, the root mean square error (RMSE) is less than 0.33 kPa, and the mean absolute error (MAE) is less than 0.19 kPa. Its performance is superior to that of purely data-driven neural networks, validating the feasibility and effectiveness of this method in practical engineering. This study provides a new, efficient, and reliable method for the consolidation analysis of plain earth dams.

Key words: plain earth dam, pore water pressure evolution, physics-informed neural network