Abstract: Aiming at the existing simulation methods of oil leakage on the sea cannot reflect the weathering and diffusion effect with the physical process and the simulation speed is too slow, a simulation method of oil leakage on the sea based on FCE-SPH was proposed. Firstly, an immiscible Cahn-Hilliard two-phase flow emulsified smooth particle hydrodynamics diffusion simulation method FCE-SPH is proposed to describe the mixing and emulsification decomposition process of sea oil spills by combining the fluctuating wind field. Secondly, an offset-attention based two-phase flow physical-aware graph attention network deep reinforcement learning network OTF-DRL is designed to accelerate fluid simulation. The offset heterogeneous graph attention module is used to enhance the represent ability of the network to obtain particle feature information. Under different working conditions, the physical process of crude oil leakage and diffusion and the oil weathering process are analyzed. The experimental results show that the proposed method can simulate the offshore oil spill diffusion effectively.