The multistage throttle orifice plate is a fluid pressure reducing device in the pipeline, its depressurization effect is affected by different parameters, such as the position and diameter of the holes in the throttle orifice plates, the distance between the throttle orifice plates. A numerical computational resource is required to achieve the flow characteristic with different parameters by computational fluid dynamics (CFD). Based on the neural network model of CFD simulation, the prediction of flow characteristic of multi-layer throttle orifice plate is realized in this paper. In the procedure, an automatic CFD simulation procedure of multilayer throttle orifice plate is constructed. Through the Latin Hypercube Designs (LHD) method, the parametric geometric model is generated. Then the flow rate of throttle orifice plate under different parameters are achieved under adaptive mesh generation technology and CFD simulation. The relationship between geometric parameters and the flow rate of throttle orifice plate is studied by multi-layer back-propagation neural network algorithm, and the flow characteristic predictive model is established. Compared with CFD simulation results, flow rate predicted by the predictive model and the maximum predictive error is 0.5%.