After the mountain oil pipeline leakage accident occurs, the simulation of the pollution spread range is of great significance in the designation of emergency rescue plans. Aiming at the existing crude oil leakage simulation methods cannot reflect the crude oil diffusion effect with the physical process and the speed of simulation is too slow, a leakage simulation method combined with the oil surface diffusion Fay model is investigated. Firstly, we propose a calculation model Fay-SPH combined with the smoothed particle hydrodynamics (SPH) method and the Fay fluid diffusion equation based on spatial interaction semantic constraints. Secondly, in order to make the simulation results quickly presented, a Fay-constrained continuous convolutional network FayNet is designed to speed up the simulation, and a lightweight attention module FGAB and a triple attention module FTAB are proposed to enhance the network's ability to extract particle feature information. Finally, the position loss and velocity loss are fused to ensure the consistency of the simulation results in detail. Based on different landform types and different topography, the results of simulation experiment on the mountain area data of the oil pipelines collected by us show that our methods can implement the efficient simulation of crude oil leakage.