摘要
为了克服传统CFD计算需要耗费大量的计算时间与成本的缺陷,提出了一种基于深度学习的非定常周期性流场的预测框架,可以实时生成给定状态的高可信度的流场结果。将条件生成对抗网络与卷积神经网络相结合,改进条件生成对抗网络对生成样本的约束方法,建立了基于深度学习策略采用改进的回归生成对抗网络模型,并与常规的条件生成对抗网络模型的预测结果进行对比。研究表明,基于改进的回归生成对抗网络的深度学习策略能准确预测出指定时刻的流场变量,且总时长比CFD数值模拟减少至少1个量级。
In order to overcome the shortages of the computationally expensive and timeconsuming iterative process in traditional CFD simulation,a framework based on the deep learning to predict periodic unsteady flow field is proposed,which can accurately predict real-time complex vortex flow state at different moments.The conditional generative adversarial network and convolutional neural network are combined to improve the conditional constraint method from conditional generative adversarial network.The improved regression generative adversarial network based on the deep learning is proposed.The two scenarios of conditional generative adversarial network and regression generative adversarial network are tested and compared via giving different periodic moments to predict the corresponding flow field variables.The final results demonstrate that regression generative adversarial network can estimate complex flow fields,and is faster than traditional CFD simulation over one order of magnitudes.
引文
[1]赵松原,黄明恪.模拟退火算法和POD降阶模态计算在翼型反设计中的应用[J].空气动力学学报,2007,25(2):236-240.ZHAO S Y,HUANG M K.Application of simulated annealing method and reduced order models based on POD to airfoil inverse design problems[J].ACTA Aerodynamica Sinica,2007,25(2):236-240.(in Chinese)
[2]刘浩,徐敏,叶茂.基于特征正交分解的跨声速流场重构和翼型反设计方法研究[J].空气动力学学报,2012,30(4):539-545.LIU H,XU M,YE M.Investigations of transonic flow field reconstruction and inverse airfoil design based on proper orthogonal decomposition ROM[J].Acta Aerodynamica Sinica,2012,30(4):539-545.(in Chinese)
[3] WAIBEL A,HANAZAWA T,HINTON G E,et al.Phoneme recognition using time-delay neural networks[J].IEEE Transaction on Acoustics Speech and Signal Processing,1989,37(3):328-339.
[4] LING J, KURZAWSKI A, TEMPLETON J.Reynolds averaged turbulence modelling using deep neural networks with embedded invariance[J].Journal of Fluid Mechanics,2016,807:155-166.
[5] LECUN Y,LENET-5,convolutional neural networks[EB/OL].http://yannl.ecuu.com/exdb/lenet/index.html
[6] MIYANAWALA T P,JAIMAN R K.An efficient deep learning technique for the Navier-Stokes equations:Application to unsteady wake flow dynamics[J].arXiv preprint arXiv:1710.09099,2017.
[7] LEE S,YOU D.Prediction of laminar vortex shedding over a cylinder using deep learning[J].Journal of Fluid Mechanics,2017.
[8] LEE S,YOU D.Data-driven prediction of unsteady flow fields over a circular cylinder using deep learning[J].Journal of Computational Physics,2018.
[9] BAZRAFKAN S,CORCORAN P.Versatile auxiliary regressor with generative adversarial network(VAR+GAN)[J].arXiv preprint arXiv:1805.10864,2018.
[10]GOODFELLOW I,POUGET-ABADIE J,MIRZA M,et al.Generative adversarialnets[C]//AdvancesinNeural Information Processing Systems,2014:2672-2680.
[11]VINCENT P,LAROCHELLE H,LAJOIE I,et al.Stacked denoising autoencoders:Learning useful representations in a deep network with a local denoising criterion[J].Journal of Machine Learning Research,2010,11:3371-3408.
[12]ZHAO J,MATHIEU M,LECUN Y.Energy-based generative adversarial network[C]//ICLR 2017.
[13]COX J S,BRENTNER K S,RUMSEY C L.Computation of vortex shedding and radiated sound for a circular cylinder:Subcritical to transcritical Reynolds numbers[J].Theoretical and Computational Fluid Dynamics,1998,12(4):233-253.
[14]RUMSEY C L,BIEDRON R T,THOMAS J L.CFL3D:Its history and some recent applications[R].NASA TM-112861.
[15]BOTTOU L.Large-scale machine learning with stochastic gradient descent[C]//Proceedings of COMPSTAT,2010:177-186.
[16]CLEVERT D A,UNTERTHINER T,HOCHREITER S.Fast and accurate deep network learning by exponential linear units(elus)[C]//ICLR 2016.
[17]NAIR V, HINTON G E.Rectified linear units improve restricted Boltzmann machines[C]//Proceedings of the 27th International Conference on Machine Learning(ICML-10).2010:807-814.
[18]FAN E.Extended tanh-function method and its applications tononlinear equations[J].Physics Letters A,2000,277(4-5):212-218.
[19]KINGMA D P,BA J.A method for stochastic optimization[C]//Proceedings of the 3rd Inernational Conference on Learning Representatims.