Subsurface fluid stratified sampling techniques have important applications in pollution monitoring in agriculture, industry and mining. The traditional sampling device has problems such as difficulty in going down the well, limited sampling depth, and silt clogging. This study proposes a new strategy to optimize the design of subsurface fluid stratified sampling device based on intelligent algorithm. The basic framework of fluid simulation is constructed by using the Navier-Stokes system of equations and the Eulerian mesh method, and a size function based on the combination of fluid surface distance and solid surface distance is designed to realize adaptive stratified sampling. The internal flow channel design of the sampling device is optimized by particle classification, splitting and merging operations. The results show that: different screen porosity (0.4-0.8) has a significant effect on the total pressure of the internal flow channel of the sampler, and the total pressure at the front end of the screen increases with the increase of the pressure drop; the maximum difference between the dryness of the split fluid and the inlet dryness is not more than 2.5%, which proves that the split fluid has a good representativeness; the split coefficients of the liquid phase and the gas phase are larger than the theoretical values in general, and they are gradually close to and tend to be stabilized with the increase of the discounted flow rate of the liquid phase; When the oil content exceeds 50%, the gas-phase partition coefficient decreases about 8%, and the average uncertainty of the total flow rate does not exceed 3.55%. The shallow U-tube layered sampling device developed based on the research results has been successfully applied in several oilfield CCS projects, realizing integrated water and gas sampling and one-hole multilayer sampling functions, which significantly improves the efficiency and accuracy of subsurface fluid sampling.