计算机科学

计算机科学 ›› 2025, Vol. 52 ›› Issue (6): 88-95.doi: 10.11896/jsjkx.241100026

• 高性能计算 • 上一篇    下一篇

基于高斯混合判别的半监督学习流场预测方法

王枭1,2, 李冠雄3, 李娜1,2, 袁东风4,5   

  1. 1 齐鲁理工学院智能制造与控制工程学院 济南 250200
    2 齐鲁理工学院山东省工业大数据与智能制造重点实验室 济南 250200
    3 四川大学工程数值模拟基础算法与模型全国重点实验室 成都 610065
    4 山东大学深圳研究院 深圳 518057
    5 山东大学齐鲁交通学院 济南 250002
  • 收稿日期:2024-11-04 修回日期:2025-01-06 出版日期:2025-06-15 发布日期:2025-06-11
  • 通讯作者: 王枭(wangxiao940218@163.com)
  • 基金资助:
    国家重点研发计划(2024YFB3312302);齐鲁理工学院人才项目(QIT24TP027);国家自然科学基金(62271288)

Semi-supervised Learning Flow Field Prediction Method Based on Gaussian Mixture Discrimination

WANG Xiao1,2, LI Guanxiong3, LI Na1,2, YUAN Dongfeng4,5   

  1. 1 College of Intelligent Manufacturing and Control Engineering,Qilu Institute of Technology,Jinan 250200,China
    2 Shandong Provincal Key Laboratory of Industrial Big Data and Intelligent Manufacturing,Qilu Institute of Technology,Jinan 250200,China
    3 National key Laboratory of fundamental Algorithms and Models for Engineering Simulation,Sichuan University,Chengdu 610065,China
    4 Shenzhen Research Institute of Shandong University,Shandong University,Shenzhen 518057,China
    5 College of Qilu Transportation,Shandong University,Jinan 250002,China
  • Received:2024-11-04 Revised:2025-01-06 Online:2025-06-15 Published:2025-06-11
  • About author:WANG Xiao,born in 1994,Ph.D,asso-ciate professor.His main research in-terests include deep learning,data mining,signal processing and AI for Science.
  • Supported by:
    National Key R&D Program of China(2024YFB3312302),Qilu Institute of Technology for Talent Project(QIT24TP027) and National Natural Science Foundation of China(62271288).

PDF (PC)

143

摘要： 深度学习在飞机设计中备受瞩目,特别是在AI for Science的推动下,基于神经网络的数据驱动方法在翼型流场预测方面取得了显著成功。然而,在标注数据有限的情况下,这些方法往往表现欠佳。针对该问题,提出了一种名为 Semi-Flow 的半监督学习方法,用于翼型流场预测。Semi-Flow 利用神经网络的损失记忆特性,根据损失函数值,将伪标签数据分为简单和困难两个子集。这种聚类方法基于高斯混合模型,将损失函数结合数据损失和辅助物理监督,确保模型结果符合气动特性和数据约束。在数据选择过程中,选择两个模型共同的简单样本作为训练样本,避免噪声样本的影响。训练过程首先对标注样本进行几轮热身训练,然后逐步添加经过过滤的简单样本。实验结果表明,Semi-Flow 方法在标记数据有限的情况下相比于仅基于少量标记数据训练表现优异,总体预测性能提升了近30%。消融研究和定性结果验证了其有效性。Semi-Flow展示了 AI for Science 的潜力,通过减少对大量标注数据的依赖,为流场预测提供了有前景的方法。

关键词: 深度学习, 面向科学任务的人工智能, 半监督学习, 流场预测

Abstract: Deep learning has garnered significant attention in aircraft design,particularly with the advancements driven by AI for Science.Data-driven methods based on neural networks have achieved remarkable success in airfoil flow field prediction.How-ever,these methods often underperform when labeled data is limited.This paper proposes a semi-supervised learning(SSL) me-thod named Semi-Flow for airfoil flow field prediction.Semi-Flow leverages the memory properties of neural network loss to classi-fy pseudo-labeled data into easy and hard subsets based on loss function values.This clustering method is based on Gaussian mixture model(GMM).The loss function combines data loss with auxiliary physical supervision,ensuring the model's outputs conform to aerodynamic properties and data constraints.During the data selection process,the easy samples common to both mo-dels are chosen as training samples,thereby avoiding the impact of noisy samples.The training process starts with several rounds of warm-up training on the labeled samples,followed by the gradual inclusion of filtered easy samples.Experimental results demonstrate that the Semi-Flow method significantly outperforms models trained solely on limited labeled data,with an overall predictionperformance improvement of nearly 30%.Ablation studies and qualitative results further validate the effectiveness of the proposed method.Semi-Flow exemplifies the potential of AI for Science,offering a promising approach to flow field prediction by reducing the dependency on large amounts of labeled data.

Key words: Deep learning, AI for science, Semi-supervised learning, Flow field prediction

中图分类号: 

  • TP391
[1]ANDERSON J D,WENDT J.Computational fluid dynamics:volume 206 [M].Springer,1995.
[2]VAN NOORDEN R,PERKEL J M.AI and Science:what1,600researchers think[J].Nature,2023,621(7980):672-675.
[3]XU Z M.Ai4science:Neural networks for molecular propertyprediction[EB/OL].https://zhiming-xu.github.io/files/Molecular_GNN.pdf.
[4]RAISSI M,PERDIKARIS P,KARNIADAKIS G E.Physics-informed neural networks:A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations[J].Journal of Computational Physics,2019,378:686-707.
[5]LU L,JIN P,PANG G,et al.Learning nonlinear operators via deeponet based on the universal approximation theorem of ope-rators[J].Nature Machine Intelligence,2021,3(3):218-229.
[6]RAZAVIAN A S,AZIZPOUR H,SULLIVAN J,et al.CNNfeatures off-the-shelf:an astounding baseline for recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops.2014:806-813.
[7]SHI W,RAJKUMAR R.Point-GNN:Graph neural network for 3d object detection in a point cloud[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2020:1711-1719.
[8]GUO X,LI W,IORIO F.Convolutional neural networks forsteady flow approximation[C]//Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining.2016:481-490.
[9]PARK J J,FLORENCE P,STRAUB J,et al.Deepsdf:Learning continuous signed distance functions for shape representation[C]//Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition.2019:165-174.
[10]DENG K,CHEN H,ZHANG Y.Flow structure oriented opti-mization aided by deep neural network[C]//10th International Conference on Computational Fluid Dynamics(ICCFD10).2018.
[11]THUEREY N,WEISSENOW K,PRANTL L,et al.Deep learning methods for reynolds-averaged navier-stokes simulations of airfoil flows[J].AIAA Journal,2020,58(1):25-36.
[12]RONNEBERGER O,FISCHER P,BROX T.U-Net:Convolu-tional networks for biomedical image segmentation[C]//International Conference on Medical Image Computing and Computer-Assisted Intervention.Springer,2015:234-241.
[13]XU Z H,CHEN X L,LI J,et al.Graph Hypersphere Prototype Network for Semi-supervised Few-shot Node Classification.Journal of Chinese Computer Systems,2025,46(3):542-551.
[14]SANCHEZ-GONZALEZ A,GODWIN J,PFAFF T,et al.Learning to simulate complex physics with graph networks[C]//International Conference on Machine Learning.PMLR,2020:8459-8468.
[15]CHEN J,HACHEM E,VIQUERAT J.Graph neural networks for laminar flow prediction around random two-dimensional shapes[J].Physics of Fluids,2021,33(12):123607.
[16]YANG Z,DONG Y,DENG X,et al.Amgnet:Multi-scale graph neural networks for flow field prediction[J].Connection Science,2022,34(1):2500-2519.
[17]WANDEL N,WEINMANN M,KLEIN R.Learning incom-pressible fluid dynamics from scratch-towards fast,differentiable fluid models that generalize[J].arXiv:2006.08762,2020.
[18]TOME M F,MCKEE S.Gensmac:A computational marker and cell method for free surface flows in general domains[J].Journal of Computational Physics,1994,110(1):171-186.
[19]GAO H,SUN L,WANG J X.Phygeonet:Physics-informed geometryadaptive convolutional neural networks for solving parameterized steadystate pdes on irregular domain[J].Journal of Computational Physics,2021,428:110079.
[20]LU L,JIN P Z,PANG G F,et al.Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators[J].Nature Machine Intelligence,2021,3:218-229.
[21]WANG J,HE C,LI R,et al.Flow field prediction of supercritical airfoils via variational autoencoder based deep learning framework[J].Physics of Fluids,2021,33(8):086108.
[22]JIANG J,LI G,JIANG Y,et al.Transcfd:A transformer-based decoder for flow field prediction[J].Engineering Applications of Artificial Intelligence,2023,123:106340.
[23]VASWANI A,SHAZEER N,PARMAR N,et al.Attention is all you need [C]//Proceedings of the 31 International Conference on Neural Information Processing Systems.2017:6000-6010.
[24]WANG X,DONG Y,ZOU S,et al.A semi-supervised framework for computational fluid dynamics prediction[J].Applied Soft Computing,2024,154:111422.
[25]WEISS K,KHOSHGOFTAAR T M,WANG D.A survey oftransfer learning[J].Journal of Big data,2016,3(1):1-40.
[26]LAI W S,HUANG J B,YANG M H.Semi-supervised learning for optical flow with generative adversarial networks[C]//Advances in Neural Information Processing Systems.2017.
[27]WANG X,JIANG Y,LI G,et al.Sag-flownet:self-attentiongenerative network for airfoil flow field prediction[J].Soft Computing,2024,28:7417-7437.
[28]WANG X,ZOU S,JIANG Y,et al.Swin-FlowNet:Flow field oriented optimization aided by a CNN and Swin-Transformer based model[J].Journal of Computational Science,2023,72:102121.
[29]WU H Z,LIU X J,AN W,et al.A generative deep learning framework for airfoil flow field prediction with sparse data[J].Chinese Journal of Aeronautics,2022,35(1):470-484.
[30]REYNOLDS D A.Gaussian mixture models[M]//Encyclopedia of Biometrics.Boston,MA:Springer,2009:659-663.
[31]YIN X,CHEN S,HU E.Regularized soft k-means for discriminant analysis[J].Neurocomputing,2013,103:29-42.
[32]ROGERS D F.An introduction to NUBRBS:with historical perspective[M].Morgan Kaufmann,2001.
[33]BLASINGAME T,JOHNSTON J,LEE W.Type-curve analysis using the pressure integral method[C]//SPE Western Regional Meeting.SPE,1989.
[34]LOH W L.On latin hypercube sampling[J].The Annals of Statistics,1996,24(5):2058-2080.
[35]LOSHCHILOV I,HUTTER F.Decoupled weight decay regularization[J].arXiv:1711.05101,2017.
Viewed
Full text


Abstract

Cited
  Shared   
  Discussed   
No Suggested Reading articles found!
渝ICP备16013121号-4
地址：重庆市渝北区洪湖西路18号    邮编：401121  
电话：023-63500828（编辑部） 023-67039612（会议/专题） 023-67039623（财务/发票）
E-mail：jsjkx12@163.com
本系统由北京玛格泰克科技发展有限公司设计开发