|
耦合流体动力学方程的神经网络模型研究
|
Abstract:
在流体力学中应用深度学习的技术成为研究热点,但是目前使用的纯粹数据驱动的神经网络模型,不具备物理知识定律的可解释性,而且作为预测模型时,其预测准确度较差。基于此,提出了耦合物理定律的神经网络模型,以层流二维圆柱绕流为例,实现了对流体动力学方程(N-S方程)的耦合与近似求解,并对模型的短时流场预测能力进行了验证。结果表明:该模型可以使用少量的训练数据对控制方程进行近似求解,对相应已知时刻的流场进行重建;进行未知时刻流场的预测时,与相同条件下纯数据驱动的神经网络流场预测模型相比,该模型具有更小的预测误差。
In recent years, the application of deep learning technology in fluid mechanics has become a re-search hotspot. However, currently pure data-driven neural network models do not have the inter-pretability of the laws of physical knowledge, and when used as a prediction model, its prediction accuracy is poor. Based on this, a neural network model coupled with the laws of physics was de-veloped. Taking laminar two-dimensional cylindrical flow as an example, the approximate solution of the N-S equation was achieved, and the short-term flow field prediction ability of the model was verified. The results show that the model can use a small amount of training data to approximate the governing equations and reconstruct the corresponding flow field. When predicting the flow field, the model has smaller prediction error than pure data-driven neural network model under the same conditions.
[1] | LeCun, Y., Bengio, Y. and Hinton, G. (2015) Deep Learning. Nature, 521, 436. https://doi.org/10.1038/nature14539 |
[2] | Bishop, C.M. (1995) Neural Networks for Pattern Recognition. Oxford University Press, Oxford.
https://doi.org/10.1201/9781420050646.ptb6 |
[3] | Manning, C.D., Manning, C.D. and Schütze, H. (1999) Founda-tions of Statistical Natural Language Processing. MIT Press, Cambridge. |
[4] | Egmont-Petersen, M., de Ridder, D. and Handels, H. (2002) Image Processing with Neural Networks—A Review. Pattern Recognition, 35, 2279-2301. https://doi.org/10.1016/S0031-3203(01)00178-9 |
[5] | 胡铁松, 周彦辰, 王先甲. 有一定物理基础的核素浓度预测神经网络模型[J]. 系统工程理论与实践, 2016, 36(1): 263-272. |
[6] | Kutz, J.N. (2017) Deep Learning in Fluid Dynamics. Journal of Fluid Mechanics, 814, 1-4.
https://doi.org/10.1017/jfm.2016.803 |
[7] | 王德明, 王莉, 张广明. 基于遗传BP神经网络的短期风速预测模型[J]. 浙江大学学报: 工学版, 2012(5): 837-841. |
[8] | 赵超, 李佳威, 伍耐明. 基于神经网络的流动预测模型研究[J]. 热能动力工程, 2017, 32(11): 8-12. |
[9] | 阮泉, 吴铁军. 结合先验知识的神经网络在生化系统建模中的应用[J]. 无锡轻工大学学报, 2001, 20(1): 55-57. |
[10] | Joerding, W.H. and Meador, J.L. (1991) Encoding a Priori Infor-mation in Feed forward Networks. Neural Networks, 4, 847-856. https://doi.org/10.1016/0893-6080(91)90063-B |
[11] | 娄海川, 苏宏业, 谢磊. 融合过程先验知识的递归神经网络模型及其应用[J]. 化工学报, 2013, 64(5): 1665-1673. |
[12] | 周彦辰, 胡铁松, 陈进, 等. 耦合动态方程的神经网络模型在水质预测中的应用[J]. 长江科学院院报, 2017, 34(9): 5-9. |
[13] | Raissi, M., Perdikaris, P. and Karniadakis, G.E. (2017) Physics Informed Deep Learning (Part I): Data-Driven Solutions of Nonlinear Partial Differential Equa-tions. |
[14] | Raissi, M., Perdikaris, P. and Karniadakis, G.E. (2017) Physics Informed Deep Learning (Part II): Da-ta-Driven Discovery of Nonlinear Partial Differential Equations. |
[15] | 吴子牛. 计算流体力学基本原理[M]. 北京: 科学出版社, 2001. |
[16] | Goodfellow, I., Bengio, Y. and Courville, A. (2016) Deep Learning. MIT Press, Cam-bridge. |
[17] | Sherwin, S.J. and Karniadakis, G.E. (2005) Spectral/HP Element Methods for Computational Fluid Dy-namics. Oxford Science Publications, Oxford, 17, 18. https://doi.org/10.1093/acprof:oso/9780198528692.001.0001 |