Physically-based simulation for oil leakage and diffusion on river using heterogeneous graph attention network.

Area of Science:

• Environmental Science
• Fluid Dynamics
• Computational Science

Background:

• Accurate simulation of oil pipeline leakage on rivers is crucial for effective emergency response.
• Existing simulation methods lack precision and are computationally intensive, hindering real-time application.
• Crude oil spills on water present complex two-phase flow dynamics involving mixing and decomposition.

Purpose of the Study:

• To develop a precise and efficient simulation method for two-phase oil and water leakage on rivers.
• To address the limitations of existing methods in terms of accuracy and computational complexity.
• To enhance the simulation's ability to capture the physical processes of oil diffusion.

Main Methods:

• Proposed a Mixture Tension Divergence-Free Smoothed Particle Hydrodynamics (MTDF-SPH) method, integrating mixture and surface tension models into divergence-free SPH.
• Developed a physics-aware heterogeneous graph attention network (PAGATNet) utilizing Attention Graph Network Blocks (AGNB) and Feature-Response Knowledge Distillation (FRKD).
• PAGATNet enhances the extraction of particle features relevant to physical properties for accelerated simulation.

Main Results:

• The MTDF-SPH method accurately simulates the mixing and decomposition of immiscible oil and water phases.
• PAGATNet significantly accelerates the leakage diffusion simulation process.
• Experimental results demonstrate superior accuracy, robustness, and effectiveness compared to state-of-the-art methods.

Conclusions:

• The proposed MTDF-SPH and PAGATNet combined approach offers a significant advancement in two-phase oil spill simulation.
• This method provides a reliable tool for predicting leakage diffusion ranges, aiding in emergency rescue plan designation.
• The study highlights the potential of integrating physics-based models with deep learning for complex environmental simulations.