Forecasting of nonlinear dynamics based on symbolic invariance

Forecasting of nonlinear dynamics based on symbolic invariance
复制标题

基于符号不变性的非线性动力学预测

DOI:
10.1016/j.cpc.2022.108382
复制
发表时间:
2022
影响因子:
6.3
通讯作者:
Sun, Hao
Sun, Hao
中科院分区:
物理与天体物理2区
文献类型:
--
作者:
Chen, Zhao;Liu, Yang;Sun, Hao

文献摘要

参考文献

相似文献

预测未知动力学在许多物理学相关学科中都有很大的兴趣。然而,数据驱动的机器学习方法受到泛化能力差的问题的困扰。为此,基于符号不变性的预测模型(即,表示内在系统机制的符号表达式/方程)。通过训练和修剪包裹在数值积分器中的符号神经网络,我们开发了一个代表进化函数的不变符号结构,从而可以很好地推广到看不见的数据。为对抗噪音影响,亦已利用非参数贝叶斯推断方法开发概率预测的算法框架。此外,为了解释从具有多个状态变量的系统中部分观察到的单变量预测,我们进一步利用延迟坐标嵌入来在更自包含的嵌入中找到部分观察系统的符号不变性。所提出的框架的性能已在合成和现实世界的非线性动力学方面得到证明,并且在短期/中期预测范围内比流行的深度学习模型具有更好的泛化能力。此外,与基于字典的符号回归方法的比较表明,当函数搜索空间巨大时,所提出的框架具有更好的性能和更有效的优化。
Forecasting unknown dynamics is of great interest across many physics-related disciplines. However, data-driven machine learning methods are bothered by the poor generalization issue. To this end, a forecasting model based on symbolic invariance (i.e., symbolic expressions/equations that represent intrinsic system mechanisms) is proposed. By training and pruning a symbolic neural network wrapped in a numerical integrator, we develop an invariant symbolic structure that represents the evolution function and thus can generalize well to unseen data. To counter noise effect, an algorithmic framework for probabilistic forecasting has also been developed by leveraging a non-parametric Bayesian inference method. Additionally, to account for univariate forecasting that is partially observed from a system with multiple state variables, we further leverage the delay coordinate embedding to find symbolic invariance of the partially observed system in a more self-contained embedding. The performance of the proposed framework has been demonstrated on both synthetic and real-world nonlinear dynamics and shown better generalization over popular deep learning models in short/medium forecasting horizons. Moreover, comparison with dictionary-based symbolic regression methods suggests better-behaved and more efficient optimization of the proposed framework when the function search space is enormous.
DOI: 10.1073/pnas.1517384113
发表时间: 2016-04-12
影响因子: 11.1
作者:
Brunton, Steven L.;Proctor, Joshua L.;Kutz, J. Nathan
通讯作者: Kutz, J. Nathan
DOI: 10.1098/rspa.2018.0305
发表时间: 2018-09
期刊: Proceedings of the Royal Society A: Mathematical, Physical and Engineering Sciences
影响因子: --
作者:
Sheng Zhang;Guang Lin
通讯作者: Sheng Zhang;Guang Lin
DOI: 10.1016/j.jcp.2018.10.045
发表时间: 2019-02-01
影响因子: 4.1
作者:
Raissi, M.;Perdikaris, P.;Karniadakis, G. E.
通讯作者: Karniadakis, G. E.
DOI: 10.1063/5.0010886
发表时间: 2020-07-01
期刊: CHAOS
影响因子: 2.9
作者:
Pan, Shaowu;Duraisamy, Karthik
通讯作者: Duraisamy, Karthik
含外生变量的向量自回归移动平均模型的快速估计方法
DOI: --
发表时间: 1983
期刊:
影响因子: --
作者:
H. Spliid
通讯作者: H. Spliid