欢迎访问过程工程学报, 今天是

过程工程学报 ›› 2024, Vol. 24 ›› Issue (8): 904-913.DOI: 10.12034/j.issn.1009-606X.224006

• 研究论文 • 上一篇    下一篇

基于人工神经网络的流化床流场演化模拟

吴雪岩1,2， 史天乐2,3， 李飞2*， 于三三1， 卢春喜3， 王维2   

  1. 1. 沈阳化工大学化学工程学院，辽宁 沈阳 110142 2. 中国科学院过程工程研究所介科学与工程全国重点实验室，北京 100190 3. 中国石油大学(北京)化学工程与环境学院，北京 102249
  • 收稿日期:2024-01-05 修回日期:2024-03-11 出版日期:2024-08-28 发布日期:2024-08-22
  • 通讯作者: 李飞 lifei@ipe.ac.cn
  • 基金资助:
    煤气化反应器的多尺度模拟方法——从微观到宏观

Simulation of flow field evolution in fluidized bed based on artificial neural network

Xueyan WU1,2,  Tianle SHI2,3,  Fei LI2*,  Sansan YU1,  Chunxi LU3,  Wei WANG2   

  1. 1. College of Chemical Engineering, Shenyang University of Chemical Technology, Shenyang, Liaoning 110142, China 2. State Key Laboratory of Mesoscience and Engineering, Institute of Process Engineering, Chinese Academy of Sciences, Beijing 100190, China 3. College of Chemical Engineering and Environment, China University of Petroleum-Beijing, Beijing 102249, China
  • Received:2024-01-05 Revised:2024-03-11 Online:2024-08-28 Published:2024-08-22

RichHTML

PDF

摘要： 计算流体动力学(CFD)是一种模拟流化床中复杂气固流动的常用方法，此方法计算效率较低，而人工神经网络(ANN)模型能克服这一缺点，实现高效计算。本工作结合CFD与人工神经网络，发展了一种快速获得流化床内流场演化的人工神经网络模型。该模型对颗粒浓度、气体压力和气固两相速度构建了不同的网络结构，以多相质点网格(MP-PIC)方法模拟流化床得到的结果作为数据集进行训练。验证结果表明，该人工神经网络模型成功实现了对流化床中的颗粒浓度、气体压力和气固两相速度的预测。在精度方面，三种网络结构模型均可准确预测一个时间步长的流场数据，在进行长时间流场预测时仍存在误差。在计算效率方面，人工神经网络模型的计算速度约为MP-PIC方法的13 000倍。

关键词: 流场演化, 人工神经网络, 多相流, 流化床, 计算流体力学

Abstract: Computational fluid dynamics (CFD) is a commonly used method to simulate complex gas-solid flow in fluidized beds. Due to the solution of partial/ordinary differential equations, the computational efficiency of this method is still low even if the coarse-grained method is used. The flow field simulation method based on data-driven artificial neural network (ANN) model can avoid the equation solving process and achieve efficient calculation. At present, researchers have applied the ANN model to the prediction of single-phase flow field, and there are only a few studies on the complete fluidized gas-solid two-phase flow field. This work combines CFD and ANN to develop an ANN based field evolution model that quickly obtains the evolution of the flow field in the fluidized bed. Compared with those complex large models, a compact network model has been developed and can be used to complete the prediction of complex two-phase flow field. The model includes different network structures for predictions of particle concentration, gas pressure, and gas-solid two-phase velocity. The results obtained by simulating the fluidized bed with the multiphase particle-in-cell (MP-PIC) method are used as data sets for training. The verification results show that the ANN model successfully realizes the prediction of particle concentration, gas pressure, and gas-solid two-phase velocity in the fluidized bed. In terms of accuracy, the ANN model can accurately predict the flow field in a time step, and there are still obvious errors in the long-term flow field prediction. In terms of computational efficiency, the calculation speed of the ANN model is about 13 000 times that of the MP-PIC method. The multi-time-step continuous prediction performance of current model gradually deteriorates with time, and further research still needs to be done to improve this issue.

Key words: Flow field evolution, Artificial neural network, Multiphase flow, Fluidized bed, Computational fluid mechanics