Abstract:

It is one of the hot topics in the field of process engineering to accurately understand and predict the complex hydrodynamics, transport phenomena, and reaction characteristics in multiphase reactors. In recent years, with the rapid development of experimental detection technology and high-performance computer, researchers can obtain high-precision multidimensional transient flow field data sets. In the last decade, machine learning, as a new interdisciplinary subject, is widely applied in data mining, image recognition, intelligent control, etc. This article summarizes several common machine learning methods, including neural network model, support vector machine model, decision tree model, clustering algorithm model, etc. Afterward, a summarization of the construction process of the machine learning model, including data set establishment, feature variable selection, algorithm framework selection, model parameter optimization, model validation, and testing, etc, is also provided. Subsequently, the application progress of machine learning assisted multiphase reactors in constitutive model construction, flow field image reconstruction, flow pattern identification, prediction and optimization of key parameters in the flow field, uncertainty analysis, and digital twin technology platform are reviewed. Finally, we analyze the challenges in the field of coupling machine learning with multiphase reactors, meanwhile, the possible development direction of machine learning in multiphase reactors is prospected.

Key words: multiphase reactor, hydrodynamics, constitutive model development, machine learning

摘要：

准确理解并精确预测多相反应器内复杂的流体力学特性、传递现象及反应特征，是过程工程领域的热点方向之一。随着试验测量技术及高性能计算机的快速发展，研究者可以获取高精度的多维瞬态流场数据集。近十年来，机器学习作为一门新兴学科，越来越广泛地应用于数据挖掘、图像识别、智能控制等领域。本文概述了几种常用的机器学习方法（包括神经网络模型、支持向量机模型、决策树模型、聚类算法模型等），总结了机器学习模型的构建过程（包括数据集的建立、特征变量的选择、算法框架的选取、模型参数的调优、模型验证与测试等），综述了机器学习辅助多相反应器中流场本构模型构建、流场图像重构、流型识别、流场关键参数预测及优化、不确定度分析、数字孪生技术平台等方面的应用进展，剖析了机器学习结合多相反应器领域所面临的挑战，展望了机器学习在多相反应器中可能有待拓展的方向。

关键词: 多相反应器, 流体力学, 本构模型构建, 机器学习