An inverse design method of pressure distribution based on machine learning technology is developed aiming at the design difficulty of non-axisymmetric Natural Laminar Flow （NLF） nacelles under the three-dimensional flow effect. The Free Form Deformation （FFD） technology is used for the parametric modeling of the nacelle. Natural transition prediction is realized by solving the RANS equation and transition model \gamma \mathrm{⁃}\overline{R{e}_{\theta t}} based on the SST （Shear Stress Transport） turbulence model. The mapping relationship between the high dimensional data set and the low dimensional hidden space variables is established with the Generative Topological Mapping （GTM） model. The global optimization algorithm efficiently optimizes in the hidden space， obtaining the aerodynamic shape of the nacelle matching the target pressure distribution， thereby realizing the inverse design of the natural laminar flow nacelle. The GTM model establishes a high-precision mapping relationship between the high and low dimensions of the data set， hence requiring no repeated calling of the CFD solver in the optimization design process， significantly improving the design efficiency. The three-dimensional optimization of the ventilation nacelle is then conducted with this method. The maximum length of the natural laminar flow area on the outer surface is 40.5% of the local chord length with the extension of 12.2%， verifying the optimization design ability of the NLF nacelle considering the three-dimensional flow effect.