Comparative prediction of pressure and velocity in 3D flow field based on neural networks

Abstract

As an important component in the coal liquefaction system, the regulating valve's flow field and pressure distribution affects the service life and working stability of the system. In order to achieve rapid prediction of the three-dimensional(3D) pressure and velocity in the regulating valve, a prediction model based on neural network was built. The hyperparameters of the model were selected and the network parameters were optimized through genetic algorithms. The training model was verified under different working conditions. The value of axial velocity and radial velocity predicted by the optimized GA-BP model are discussed. The predicted axial velocity v_x, radial velocity v_y and v_z are almost the same with the simulation results. The largest velocity located near the orifice because of the sudden decreasing flow area, and there is a local low speed area near the head of the core head. And the distribution of pressure in the valve is also predicted by this proposed GA-BP model. There is a reflux with local low pressure is located near the orifice, and the error between the simulation and predicted results is about 2 %. Furthermore, the 3D flow field in the regulating valve with higher working pressure is predicted which cannot be easily measured experimentally. The value of resultant velocity $\bar{v}$ is close to the axial velocity ${\bar{v}}_x$, the maximum value of ${\bar{v}}_x$ is about 200 m.s⁻¹ which is located near the orifice. The value of radial velocity $\left|{\bar{v}}_y\right|$ and $\left|{\bar{v}}_z\right|$ are almost the same, because the structure of the experimental valve is axisymmetric. The maximum value of $\left|{\bar{v}}_y\right|$ and $\left|{\bar{v}}_z\right|$ are 38.3 m.s⁻¹ and 36.2 m.s⁻¹ respectively. This GA-BP prediction model has a good learning effect on the characteristics of the flow field in the regulating valve, could reflect and predict the operation status of the system.