基于生成对抗网络的多相流搅拌流场研究  

Research on Mixing Flow Field of Multiphase Flow Based on Generative Adversarial Network

作　　者：宋赫轩 郑卫芳[1] 李传博[1] Song Hexuan;Zheng Weifang;Li Chuanbo(China Institute of Atomic Energy,Beijing 102413,China)

机构地区：[1]中国原子能科学研究院,北京102413

出　　处：《广东化工》2025年第1期47-50,共4页Guangdong Chemical Industry

摘　　要：多相流搅拌流场作为一种复杂的流场状态常常难以拟合,但其作为一个常见的流体状态广泛存在于化工、冶金、制药等诸多工艺生产中。神经网络具备强大的拟合能力刚好可以对传统方法难以拟合的复杂流场状态进行拟合。本文采用新型的WGAN-GP网络架构,成功对传统BP神经网络线性结构无法拟合的多相流搅拌流场进行拟合,最终流场参数相对误差仅为2%左右。证明了BP神经网络具备充分的针对多相流搅拌流场的拟合能力,即简单的网络结构依然具备较强的拟合能力,为之后的复杂流场拟合提供了一种计算速度快、消耗资源少的计算思路。As a complicated flow field,it is always difficult to fit,but as a common fluid state,it is widely used in chemical,metallurgical,pharmaceutical and other processes.The neural network has strong fitting ability and can fit the complicated flow field state which is difficult to fit by traditional methods.In this paper,a new WGAN-GP network architecture is used to successfully fit the stirred flow field of multiphase flow,which cannot be fitted by the linear structure of the traditional BP neural network.The relative error of the final flow field parameters is only about 2%.It is proved that the BP neural network has sufficient fitting ability for the stirred flow field of multiphase flow,that is,the simple network structure still has strong fitting ability,which provides a calculation idea for the complex flow field fitting with fast computing speed and less resource consumption.

关 键 词：多相流 搅拌流场 BP神经网络 生成对抗网络 

分 类 号：TB126[理学—工程力学]