基于图神经网络的铅铋快堆上腔室三维热分层现象非线性降阶分析

曾付林; 赵鹏程; 李玲莉; 刘紫静; 李卫

doi:10.13832/j.jnpe.2024.050044

基于图神经网络的铅铋快堆上腔室三维热分层现象非线性降阶分析

doi: 10.13832/j.jnpe.2024.050044

1.
南华大学核科学技术学院，湖南衡阳，421001
2.
国立核能研究大学莫斯科工程物理学院，莫斯科，115409
3.
南华大学资源环境与安全工程学院，湖南衡阳，421001

基金项目: 核反应堆系统设计技术重点实验室运行基金（KFKT-05-FW-HT-20220014）

详细信息

作者简介:
曾付林（2002—），男，硕士研究生，现主要从事先进反应堆物理与热工安全分析研究，E-mail: 3077819552@qq.com

通讯作者:
赵鹏程，E-mail: pengcheng.zhao@usc.edu.cn

中图分类号: TL33
计量
- 文章访问数: 11
- HTML全文浏览量: 5
- PDF下载量: 2
- 被引次数: 0
出版历程
- 收稿日期: 2024-05-27
- 修回日期: 2024-07-12
- 网络出版日期: 2025-06-09
- 刊出日期: 2025-06-09

Nonlinear Reduced-order Analysis of Three-dimensional Thermal Stratification in the Upper Plenum of Lead-bismuth Cooled Fast Reactor Based on Graph Neural Network

1.
School of Nuclear Science and Technology, University of South China, Hengyang, Hunan, 421001, China
2.
Moscow Engineering Physics Institute, National Research Nuclear University, Moscow, 115409, Russia
3.
School of Resources Environment and Safety Engineering, University of South China, Hengyang, Hunan, 421001, China

摘要

摘要: 铅铋快堆上腔室的热分层现象对堆内构件的结构安全性和系统余热排出能力具有重要影响，需重点分析。本文首先基于计算流体动力学（CFD）程序FLUENT得到铅铋快堆上腔室热分层现象的高精度全阶快照，然后利用图神经网络（GNN）构建的图自编码器（GAE）对快照进行非线性降阶，并将非线性降阶后的重构结果与本征正交分解（POD）的线性降阶结果进行对比分析，最后通过结合多层感知机对热分层快照开展在线状态识别与预测分析。结果表明，由于GNN具有高度的非线性，使其在大规模CFD数据的非线性降阶方面具备独特优势，其1阶模态重构精度与POD的30~50阶基函数的重构精度相当；在在线阶段，以472 ms的时长即可完成对热分层快照的特征识别和预测，且预测精度与直接重构精度相近。相关研究成果可为铅铋快堆热分层现象演化机理分析及后果预测提供新的分析方法支撑。
- 铅铋快堆 /
- 热分层 /
- 非线性降阶 /
- 图神经网络（GNN） /
- 本征正交分解（POD）
Abstract: The thermal stratification phenomenon in the upper plenum of the lead-bismuth fast reactor has significant implications for the safety of internal components and the ability to remove residual heat. This paper focuses on analyzing this phenomenon. Firstly, high-precision full-order snapshots of the thermal stratification in the upper plenum of the lead-bismuth fast reactor are obtained using the computational fluid dynamics software FLUENT. Then, the graph neural network (GNN) is employed to construct a graph autoencoder (GAE) for nonlinear order reduction of the snapshots, and the reduced reconstruction results are compared with the linear reduction results obtained using Proper orthogonal decomposition (POD). Finally, a multilayer perceptron is used for online state recognition and predictive analysis of the thermal stratification snapshots. The research results demonstrate that the graph neural network, due to its high level of nonlinearity and its inherent advantage in nonlinear order reduction of large-scale CFD data, achieves comparable first-order modal reconstruction accuracy to that of POD with 30~50 basis functions. During the online process, the feature recognition and prediction of thermal stratification snapshots can be completed within a duration of 472 ms, with accuracy similar to the reconstruction accuracy. The related research results can provide a new analytical method support for the evolution mechanism analysis and consequence prediction of thermal stratification phenomenon in lead-bismuth fast reactors.
- Lead-bismuth cooled fast reactor /
- Thermal stratification /
- Nonlinear reduced-order model /
- Graph neural network (GNN) /
- Proper orthogonal decomposition (POD)

HTML全文

图 1 GNN特征聚合过程

Figure 1. Feature Aggregation Process in Graph Neural Networks

下载: 全尺寸图片幻灯片

图 2 GAE架构

Figure 2. GAE Architecture

下载: 全尺寸图片幻灯片

图 3 TSTF的实验区域

Figure 3. Experimental Area of the TSTF

下载: 全尺寸图片幻灯片

图 4 实验段垂直截面与模型结构

Figure 4. Vertical Section of Test Section and Model Structure

下载: 全尺寸图片幻灯片

图 5 网格结构与敏感性分析

Figure 5. Mesh Structure and Mesh-Sensitivity Analysis Results

下载: 全尺寸图片幻灯片

图 6 上腔室温度分布云图

Figure 6. Temperature Distribution of the Upper Plenum

下载: 全尺寸图片幻灯片

图 7 损失函数下降趋势

Figure 7. Decreasing Trend of the Loss Function

下载: 全尺寸图片幻灯片

图 8 测点位置温度重构对比结果

Figure 8. Temperature Reconstruction at Measurement Point

下载: 全尺寸图片幻灯片

图 9 GAE与POD重构精度对比

Figure 9. Comparison of Reconstruction Accuracy between GAE and POD

下载: 全尺寸图片幻灯片

图 10 5~45 s重构结果及误差分布云图

Figure 10. Reconstruction Results and Error Distribution Contours at 5~45 s

下载: 全尺寸图片幻灯片

图 11 低维特征随时间的变化

Figure 11. Change of Low-Dimensional Features over Time

下载: 全尺寸图片幻灯片

图 12 状态识别迭代轨迹

Figure 12. State Identification Iteration Trajectory

下载: 全尺寸图片幻灯片

图 13 第35秒和第45秒时的预测结果及误差分布云图

Figure 13. Prediction Results and Error Distribution Contours at 35th Second and 45th Second

下载: 全尺寸图片幻灯片

表 1 结构参数和实验参数数据

Table 1. Structural Parameters and Experimental Data

参数名	实验设计数据	ABTR数据	比值
上腔室高度/m	1.20	8.02	0.15
出口高度/m	0.8000	5.347	0.15
上腔室直径/m	0.3147	4.910	0.065
中间构件直径/m	0.1455	2.270	0.065
入口直径/m	0.0127(×3个)	0.3710(×3个)	0.009
出口直径/m	0.0160(×2个)
非保护性事故流量/(m³·s⁻¹)	0.001429	0.3775	0.0038
堆芯冷却剂流速/(m·s⁻¹)	3.759	0.296	12.7
腔室热流体温度/℃	300	575	0.52
腔室冷流体温度/℃	250	525	0.48
佩克莱数（Pe）	1616	12520	0.13
雷诺数（Re）	28349	788237	0.04
Ri	1616	1616	1

下载: 导出CSV

参考文献(21)

[1]	WU Z Y, LU C H, MORGAN S, et al. A status review on the thermal stratification modeling methods for Sodium-cooled Fast Reactors[J]. Progress in Nuclear Energy, 2020, 125: 103369. doi: 10.1016/j.pnucene.2020.103369
[2]	DONG Z Y, QIU H R, WANG M J, et al. Numerical simulation on the thermal stratification in the lead pool of lead-cooled fast reactor (LFR)[J]. Annals of Nuclear Energy, 2022, 174: 109176. doi: 10.1016/j.anucene.2022.109176
[3]	魏诗颖,王成龙,田文喜,等. 铅基快堆关键热工水力问题研究综述[J]. 原子能科学技术,2019, 53(2): 326-336. doi: 10.7538/yzk.2018.youxian.0335
[4]	ZHANG Y, WANG C L, LAN Z K, et al. Review of thermal-hydraulic issues and studies of lead-based fast reactors[J]. Renewable and Sustainable Energy Reviews, 2020, 120: 109625. doi: 10.1016/j.rser.2019.109625
[5]	ZWIJSEN K, UITSLAG-DOOLAARD H, ROELOFS F, et al. Fast reactor multi-scale and multi-physics modelling at NRG[J]. Nuclear Engineering and Design, 2023, 410: 112389. doi: 10.1016/j.nucengdes.2023.112389
[6]	LUO X, WANG C, ZOU Z R, et al. Development and application of a multi-physics and multi-scale coupling program for lead-cooled fast reactor[J]. Nuclear Science and Techniques, 2022, 33(2): 18. doi: 10.1007/s41365-022-01008-y
[7]	柏云清,李文博,李文东,等. 池式铅基堆热分层现象研究综述[J]. 核科学与工程,2023, 43(2): 286-297. doi: 10.3969/j.issn.0258-0918.2023.02.007
[8]	LIU D X, MARTHALER P G, CLASS A G, et al. Simulation of stratification phenomena with a spectral Fourier–Chebyshev method on the background of a lead-cooled fast reactor[J]. Annals of Nuclear Energy, 2023, 192: 109931. doi: 10.1016/j.anucene.2023.109931
[9]	HU R. Development of a reduced-order three-dimensional flow model for thermal mixing and stratification simulation during reactor transients: AC02-06CH11357[R]. Argonne: Argonne National Lab., 2017.
[10]	ZOU L, NUNEZ D, HU R. Development and validation of SAM multi-dimensional flow model for thermal mixing and stratification modeling: ANL-NSE-20/19[R]. Argonne: Argonne National Lab., 2020.
[11]	HU R. Progress report on SAM reduced-order model development for thermal stratification and mixing during reactor transients: ANL/NE-17/29[R]. Argonne: Argonne National Lab., 2017.
[12]	HU R. Three-dimensional flow model development for thermal mixing and stratification modeling in reactor system transients analyses[J]. Nuclear Engineering and Design, 2019, 345: 209-215. doi: 10.1016/j.nucengdes.2019.02.018
[13]	LORENZI S, CAMMI A, LUZZI L, et al. A reduced order model for investigating the dynamics of the Gen-IV LFR coolant pool[J]. Applied Mathematical Modelling, 2017, 46: 263-284. doi: 10.1016/j.apm.2017.01.066
[14]	杨涛,赵鹏程,赵亚楠,等. 基于CFD的铅铋快堆上腔室降阶热分层模型开发[J]. 核动力工程,2023, 44(2): 48-53.
[15]	YANG T, ZHAO P C, ZHAO Y N, et al. Development of reduced-order thermal stratification model for upper plenum of a lead–bismuth fast reactor based on CFD[J]. Nuclear Engineering and Technology, 2023, 55(8): 2835-2843. doi: 10.1016/j.net.2023.05.002
[16]	FUKAMI K, NAKAMURA T, FUKAGATA K. Convolutional neural network based hierarchical autoencoder for nonlinear mode decomposition of fluid field data[J]. Physics of Fluids, 2020, 32(9): 095110. doi: 10.1063/5.0020721
[17]	HU J W, ZHANG W W. Mesh-Conv: convolution operator with mesh resolution independence for flow field modeling[J]. Journal of Computational Physics, 2022, 452: 110896. doi: 10.1016/j.jcp.2021.110896
[18]	KASHEFI A, REMPE D, GUIBAS L J. A point-cloud deep learning framework for prediction of fluid flow fields on irregular geometries[J]. Physics of Fluids, 2021, 33(2): 027104. doi: 10.1063/5.0033376
[19]	BARWEY S, SHANKAR V, VISWANATHAN V, et al. Multiscale graph neural network autoencoders for interpretable scientific machine learning[J]. Journal of Computational Physics, 2023, 495: 112537. doi: 10.1016/j.jcp.2023.112537
[20]	PICHI F, MOYA B, HESTHAVEN J S. A graph convolutional autoencoder approach to model order reduction for parametrized PDEs[J]. Journal of Computational Physics, 2024, 501: 112762. doi: 10.1016/j.jcp.2024.112762
[21]	TOMCHIK C, VEGENDLA P, BUCKNOR M, et al. Benchmark specification for select experiments conducted at the university of Wisconsin-Madison thermal stratification test facility: ANL-ART-253[R]. Argonne: Argonne National Lab., 2022.