瞬态多相流场图神经网络时空预测方法研究  

Spatiotemporal prediction method for the transient multiphase flow field via graph neural network

作　　者：郝祎琛 谢心喻 丁家琦 谢蓉[1] 王晓放[1] 刘海涛[1] HAO Yichen;XIE Xinyu;DING Jiaqi;XIE Rong;WANG Xiaofang;LIU Haitao(College of Energy and Power,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]大连理工大学能源与动力学院,辽宁大连116024

出　　处：《哈尔滨工程大学学报》2024年第9期1761-1769,共9页Journal of Harbin Engineering University

基　　金：国家重点研发计划(2020YFA0714403);国家自然科学基金青年项目(52005074);中央高校基本科研业务费(DUT19RC(3)070);国家自然科学基金面上项目(52375231);辽宁省自然科学基金面上项目(2022-MS-135).

摘　　要：为实现对大型能源化工装备(如循环流化床)内瞬态多相流场的快速时空建模和预测,本文采用基于网格图神经网络的深度学习模型,针对循环流化床非结构化时变数值仿真数据,建立离散相体积分数的时空耦合预测器。该模型有效捕捉了反应器的时空多尺度特性,能高效地进行多相流场时空耦合动态预测,结果表明:速度远超传统数值仿真,加速比可接近500。This study aims to rapidly build a spatiotemporal model and predict transient multiphase flow fields within large-energy and chemical equipment(e.g.,circulating fluidized beds).Herein,a deep learning model based on the graph neural network was developed.It established a spatiotemporal predictor for discrete phase volume frac-tions with numerically simulated unstructured time-varying data of a circulating fluidized bed.The model success-fully captured the multiscale spatiotemporal features of the fluidized bed,achieving high-efficiency dynamic predic-tion of the spatiotemporal coupling of the multiphase flow field.The result showed that,compared with traditional numerical simulation,the data-driven model ran significantly faster,with a speedup ratio close to 500.

关 键 词：深度学习 图神经网络 循环流化床 非结构化网格 瞬态流场 多相流 时空预测 多尺度特征 

分 类 号：TK12[动力工程及工程热物理—工程热物理]