物理驱动机器学习是一种将物理原理融入机器学习框架的前沿方法。通过引入物理知识，该方法旨在使模型更为贴合实际世界的物理规律和约束，以提高模型在学习过程中对数据本质特征的准确捕捉。本研究使用了一种以支持向量机为基础的物理驱动方法，用于精确计算结构的动力响应。该算法通过最小化多输出最小二乘支持向量机的目标函数，实现了对回归模型参数的精准拟合。同时，通过在特征空间中引入系统动态平衡方程和初始条件的物理约束，无需事先训练数据即可有效计算结构的动力响应。随后开展在地震动荷载作用下的单自由度体系和二层剪切框架多自由度体系的动力响应，并将所用方法与传统方法的结果进行了对比。分析结果表明，提出的物理驱动机器学习方法在精度和大时间步长性能方面均显著优于传统方法。
Abstract
Physics-informed machine learning is a cutting-edge method that integrates physical principles into the machine learning framework. By incorporating physical knowledge, this approach aims to align the model more closely with the actual physical laws and constraints of the real world, thereby enhancing the model's precision in capturing the intrinsic characteristics of data during the learning process. This research utilized a physics-guided methodology based on support vector machines to precisely compute the dynamic response of structures. This algorithm finely tunes the regression model parameters by minimizing the objective function of a multi-output least squares support vector machine. Additionally, by embedding physical constraints pertaining to the system's dynamic equilibrium equations and initial conditions into the feature space, it is possible to effectively calculate the dynamic response of structures without the need for pre-training on data. Subsequently, the study explored the dynamic response of single-degree-of-freedom systems and multi-degree-of-freedom two-story shear frames under earthquake loads, comparing the methods used with conventional approaches. Analytical results showed that the proposed physics-informed machine learning method significantly surpasses traditional techniques in terms of accuracy and performance over large time steps.
关键词
机器学习 /
支持向量机 /
物理驱动 /
无标记数据 /
结构动力响应分析
{{custom_keyword}} /
Key words
machine learning /
support vector machine /
physics driven /
unlabeled data /
structural dynamics computation
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] NEWMARK N M. A method of computation for structural dynamics [J]. Journal of the engineering mechanics division, 1959, 85(3): 67-94.
[2] BATHE K, WILSON E. Stability and accuracy analysis of direct integration methods [J]. Earthquake Engineering & Structural Dynamics, 1972, 1(3): 283-91.
[3] CHOPRA A K. Dynamics of Structures: Theory and Applications to Earthquake Engineering, Prentice Hall [J]. Inc, Upper Saddle River, NJ, 1995.
[4] WANG N, CHANG H, ZHANG D. Deep‐learning‐based inverse modeling approaches: A subsurface flow example [J]. Journal of Geophysical Research: Solid Earth, 2021, 126(2): e2020JB020549.
[5] XU H, CHANG H, ZHANG D. DL-PDE: Deep-learning based data-driven discovery of partial differential equations from discrete and noisy data [J]. arXiv preprint arXiv:190804463, 2019.
[6] RAISSI M, KARNIADAKIS G E. Hidden physics models: Machine learning of nonlinear partial differential equations [J]. Journal of Computational Physics, 2018, 357: 125-41.
[7] RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Physics informed deep learning (part i): Data-driven solutions of nonlinear partial differential equations. arXiv 2017 [J]. arXiv preprint arXiv:171110561.
[8] RAISSI M, PERDIKARIS P, KARNIADAKIS G E. Numerical Gaussian processes for time-dependent and nonlinear partial differential equations [J]. SIAM Journal on Scientific Computing, 2018, 40(1): A172-A98.
[9] 查文舒, 李道伦, 沈路航, et al. 基于神经网络的偏微分方程求解方法研究综述 [J]. 力学学报, 2022, 54(3): 543-56.
CHA Wen-shu, LI Dao-lun, SHEN Lu-hang. Review of neural network-based methods for solving partial differential equations [J]. Chinese Journal of Theoretical and Applied Mechanics, 2022, 54(3): 543-556
[10] ZHANG Y, BURTON H V, SUN H, SHOKRABADI M. A machine learning framework for assessing post-earthquake structural safety [J]. Structural safety, 2018, 72: 1-16.
[11] YINFENG D, YINGMIN L, MING L, MINGKUI X. Nonlinear structural response prediction based on support vector machines [J]. Journal of Sound and Vibration, 2008, 311(3-5): 886-97.
[12] WU R-T, JAHANSHAHI M R. Deep convolutional neural network for structural dynamic response estimation and system identification [J]. Journal of Engineering Mechanics, 2019, 145(1): 04018125.
[13] 赵唯以, 陈沛涵. 基于机器学习的单钢板混凝土组合板冲击响应预测及优化 [J]. 振动与冲击, 2023, 42(08): 28-37.
Zhao Wei-yi, Chen Pei-han. Impact response prediction and optimization of half steel-concrete composite slabs based on machine learning [J]. Journal of vibration and shock, 2023, 42(08): 28-37.
[14] YU B. The deep Ritz method: a deep learning-based numerical algorithm for solving variational problems [J]. Communications in Mathematics and Statistics, 2018, 6(1): 1-12.
[15] 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.
[16] SIRIGNANO J, SPILIOPOULOS K. DGM: A deep learning algorithm for solving partial differential equations [J]. Journal of computational physics, 2018, 375: 1339-64.
[17] YU Y, YAO H, LIU Y. Structural dynamics simulation using a novel physics-guided machine learning method [J]. Engineering Applications of Artificial Intelligence, 2020, 96: 103947.
[18] ZHANG R, LIU Y, SUN H. Physics-guided convolutional neural network (PhyCNN) for data-driven seismic response modeling [J]. Engineering Structures, 2020, 215: 110704.
[19] 杨雅勋, 张伟德, 于海波, et al. 基于改进支持向量机的超声参数与受载混凝土应力状态研究 [J]. 振动与冲击, 2023, 42(02): 175-81+224.
Yang Ya-xun, Zhang Wei-de, Yu Hai-bo. Analysis on the relationships between ultrasonic parameters and the stress state in loaded concrete based on improved support vector machines [J]. Journal of vibration and shock, 2023, 42(02): 175-81+224.
[20] MEHRKANOON S, FALCK T, SUYKENS J A. Approximate solutions to ordinary differential equations using least squares support vector machines [J]. IEEE transactions on neural networks and learning systems, 2012, 23(9): 1356-67.
[21] 古泉, 黄素蓉. OpenSees实用教程 [M]. OpenSees实用教程, 2016.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
PDF(1810 KB)
