Abstract
Once the oil pipeline leakage accident occurs on the river, the simulation of the leakage diffusion range is of great significance for the designation of emergency rescue plans. The existing methods cannot show the precise leakage diffusion process consistent with the physical law for crude oil on the river and the simulation suffers high run-time complexity. This paper proposed a two-phase leakage simulation for oil and water combined with the physical process of smoothed particle hydrodynamics (SPH) and graph attention network. A new and efficient method-Mixture Tension Divergence-Free SPH (MTDF-SPH)-that the mixture model and the surface tension model are introduced to the divergence-free smoothed particle hydrodynamics (DFSPH) for simulating the mixing and decomposition effects of immiscible phases. To further accelerate the leakage diffusion process, we design a physics-aware heterogeneous graph attention network (PAGATNet), based on Attention Graph Network Block (AGNB) and Feature-Response Knowledge Distillation (FRKD) to enhance the network's ability for extracting the particle features of physical properties. The experimental results on different test cases show the accuracy, robustness and effectiveness of our method than those of the state-of-the-art in two-phase leakage simulation of crude oil on the river.