☆ 4.7 Article

Port-metriplectic neural networks: thermodynamics-informed machine learning of complex physical systems

COMPUTATIONAL MECHANICS (2023)

相关参考文献

注意：仅列出部分参考文献，下载原文获取全部文献信息。

Review Computer Science, Interdisciplinary Applications

Thermodynamics of Learning Physical Phenomena

Elias Cueto et al.

Summary: Thermodynamics, as a higher level of physics, has the potential to aid accurate and credible predictions in machine learning. This review explores how thermodynamics provides insights in the learning process, considering factors such as scale, choice of variables, and learning techniques.

ARCHIVES OF COMPUTATIONAL METHODS IN ENGINEERING (2023)

添加到收藏夹

Article Mathematics, Interdisciplinary Applications

A thermodynamics-informed active learning approach to perception and reasoning about fluids

Beatriz Moya et al.

Summary: Learning and reasoning about physical phenomena is a challenge in robotics, and computational sciences play a key role in finding accurate methods for explaining past events and predicting future situations. This paper proposes a thermodynamics-informed active learning strategy for fluid perception and reasoning from observations. The method demonstrates the importance of physics and knowledge in data-driven modeling and adaptation to low-data regimes.

COMPUTATIONAL MECHANICS (2023)

添加到收藏夹

Article Mathematics, Applied

Pseudo-Hamiltonian neural networks with state-dependent external forces

Solve Eidnes et al.

Summary: Hybrid machine learning based on Hamiltonian formulations has been successfully demonstrated on simple mechanical systems. A pseudo-Hamiltonian formulation is introduced to generalize the Hamiltonian formulation via the port-Hamiltonian formulation, allowing pseudo-Hamiltonian neural network models to learn external forces acting on a system. Numerical results and an improved training scheme are provided for different systems.

PHYSICA D-NONLINEAR PHENOMENA (2023)

添加到收藏夹

Article Multidisciplinary Sciences

GFINNs: GENERIC formalism informed neural networks for deterministic and stochastic dynamical systems

Zhen Zhang et al.

Summary: This paper proposes the GENERIC formalism informed neural networks (GFINNs) that obey the symmetric degeneracy conditions of the GENERIC formalism. By leveraging the component-wise architecture design, GFINNs are able to learn the underlying equations and outperform existing methods in simulation problems.

PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES (2022)

添加到收藏夹

Article Chemistry, Physical

Harnessing deep reinforcement learning to construct time-dependent optimal fields for quantum control dynamics

Yuanqi Gao et al.

Summary: We propose an efficient deep reinforcement learning (DRL) approach to automatically construct optimal control fields in dynamical chemical systems. Our approach outperforms existing gradient-based methods, even for difficult-to-converge cases. We provide a detailed description of the algorithms, hyperparameters, and performance metrics for our DRL-based approach.

PHYSICAL CHEMISTRY CHEMICAL PHYSICS (2022)

添加到收藏夹

Article Computer Science, Interdisciplinary Applications

Learning non-Markovian physics from data

David Gonzalez et al.

Summary: A method for data-driven learning of physical phenomena with internal variables and inherent noise is introduced, achieving a description that complies with basic thermodynamic principles.

JOURNAL OF COMPUTATIONAL PHYSICS (2021)

添加到收藏夹

Article Computer Science, Interdisciplinary Applications

Structure-preserving neural networks

Quercus Hernandez et al.

Summary: The method developed uses feedforward neural networks to learn physical systems from data while ensuring compliance with the first and second principles of thermodynamics. By enforcing the metriplectic structure of dissipative Hamiltonian systems, it minimizes the amount of data required and naturally achieves conservation of energy and dissipation of entropy in its predictions. No prior knowledge of the system is necessary, and the method can handle both conservative and dissipative, discrete and continuous systems.

JOURNAL OF COMPUTATIONAL PHYSICS (2021)

添加到收藏夹

Article Engineering, Multidisciplinary

Deep learning of thermodynamics-aware reduced-order models from data

Quercus Hernandez et al.

Summary: The algorithm presented in this study uses sparse autoencoders to reduce the dimensionality of a physical system and predict its time evolution using deep neural networks. It ensures the conservation of total energy and entropy inequality, making it suitable for both conservative and dissipative systems. The method has been tested in examples from fluid and solid mechanics.

COMPUTER METHODS IN APPLIED MECHANICS AND ENGINEERING (2021)

添加到收藏夹

Article Multidisciplinary Sciences

Gradient and GENERIC time evolution towards reduced dynamics

M. Grmela et al.

PHILOSOPHICAL TRANSACTIONS OF THE ROYAL SOCIETY A-MATHEMATICAL PHYSICAL AND ENGINEERING SCIENCES (2020)

添加到收藏夹

Review Automation & Control Systems

Twenty years of distributed port-Hamiltonian systems: a literature review

Ramy Rashad et al.

IMA JOURNAL OF MATHEMATICAL CONTROL AND INFORMATION (2020)

添加到收藏夹

Article Computer Science, Artificial Intelligence

SympNets: Intrinsic structure-preserving symplectic networks for identifying Hamiltonian systems

Pengzhan Jin et al.

NEURAL NETWORKS (2020)

添加到收藏夹

Article Physics, Mathematical

Machine Learning and Computational Mathematics

E. Weinan

COMMUNICATIONS IN COMPUTATIONAL PHYSICS (2020)

添加到收藏夹

Article Thermodynamics

Thermodynamically consistent data-driven computational mechanics

David Gonzalez et al.

CONTINUUM MECHANICS AND THERMODYNAMICS (2019)

添加到收藏夹

Article Automation & Control Systems

Robust port-Hamiltonian representations of passive systems

Christopher A. Beattie et al.

AUTOMATICA (2019)

添加到收藏夹

Article Materials Science, Multidisciplinary

Learning Corrections for Hyperelastic Models From Data

David Gonzalez et al.

FRONTIERS IN MATERIALS (2019)

添加到收藏夹

Article Mathematics, Applied

On learning Hamiltonian systems from data

Tom Bertalan et al.

CHAOS (2019)

添加到收藏夹

Review Physics, Multidisciplinary

GENERIC guide to the multiscale dynamics and thermodynamics

Miroslav Grmela

JOURNAL OF PHYSICS COMMUNICATIONS (2018)

添加到收藏夹

Article Mathematics

A Proposal on Machine Learning via Dynamical Systems

E. Weinan

COMMUNICATIONS IN MATHEMATICS AND STATISTICS (2017)

添加到收藏夹

Article Engineering, Multidisciplinary

Thermodynamically consistent time-stepping algorithms for non-linear thermomechanical systems

Ignacio Romero

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING (2009)

添加到收藏夹

Article Physics, Fluids & Plasmas

Nonequilibrium thermodynamics for open systems

HC Öttinger

PHYSICAL REVIEW E (2006)

添加到收藏夹

Article Engineering, Electrical & Electronic

An enhanced Lagrangian neural network for the ELD problems with piecewise quadratic cost functions and nonlinear constraints

SC Lee et al.

ELECTRIC POWER SYSTEMS RESEARCH (2002)

添加到收藏夹

© Peeref 2019-2024. All rights reserved.