Deep convolutional recurrent autoencoders for learning low-dimensional feature dynamics of fluid systems
read the original abstract
Model reduction of high-dimensional dynamical systems alleviates computational burdens faced in various tasks from design optimization to model predictive control. One popular model reduction approach is based on projecting the governing equations onto a subspace spanned by basis functions obtained from the compression of a dataset of solution snapshots. However, this method is intrusive since the projection requires access to the system operators. Further, some systems may require special treatment of nonlinearities to ensure computational efficiency or additional modeling to preserve stability. In this work we propose a deep learning-based strategy for nonlinear model reduction that is inspired by projection-based model reduction where the idea is to identify some optimal low-dimensional representation and evolve it in time. Our approach constructs a modular model consisting of a deep convolutional autoencoder and a modified LSTM network. The deep convolutional autoencoder returns a low-dimensional representation in terms of coordinates on some expressive nonlinear data-supporting manifold. The dynamics on this manifold are then modeled by the modified LSTM network in a computationally efficient manner. An offline unsupervised training strategy that exploits the model modularity is also developed. We demonstrate our model on three illustrative examples each highlighting the model's performance in prediction tasks for fluid systems with large parameter-variations and its stability in long-term prediction.
This paper has not been read by Pith yet.
Forward citations
Cited by 6 Pith papers
-
Convolutional Symmetric AutoEncoders: enhancing latent stability via differential geometry
Symmetric convolutional autoencoders preserve manifold parametrization properties, yielding more accurate latent trajectories, lower reconstruction errors, and greater robustness than standard CAEs on 1D advection, Bu...
-
One-shot learning for the complex dynamical behaviors of weakly nonlinear forced oscillators
MEv-SINDy learns governing equations from one time series to predict global frequency-response curves including softening, hardening, and jumps in non-autonomous nonlinear systems.
-
TAEN: A Model-Constrained Tikhonov Autoencoder Network for Forward and Inverse Problems
TAE combines Tikhonov regularization with autoencoders and a data randomization strategy to learn forward and inverse surrogates from one sample, with linear error bounds and tests on heat inversion and Navier-Stokes ...
-
Non-intrusive Learning of Physics-Informed Spatio-temporal Surrogate for Accelerating Design
A non-intrusive framework combines Koopman autoencoders with a spatio-temporal surrogate to learn and predict physics-constrained dynamics of systems like 2D flow around a cylinder for unseen conditions.
-
A Robust SINDy Autoencoder for Noisy Dynamical System Identification
A noise-separating SINDy autoencoder recovers interpretable latent dynamics and estimates measurement noise from noisy Lorenz observations.
-
Model reduction of parametric ordinary differential equations via autoencoders: representation properties and convergence analysis
Autoencoders enable nonlinear dimensionality reduction for parametric ODEs, with analysis of exact representation properties and convergence of the reduced model to the original.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.