A comparison of neural network architectures for data-driven reduced-order modeling

The popularity of deep convolutional autoencoders (CAEs) has engendered new and effective reduced-order models (ROMs) for the simulation of large-scale dynamical systems. Despite this, it is still unknown whether deep CAEs provide superior performance over established linear techniques or other netw...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:Computer methods in applied mechanics and engineering Ročník 393; s. 114764
Hlavní autori: Gruber, Anthony, Gunzburger, Max, Ju, Lili, Wang, Zhu
Médium: Journal Article
Jazyk:English
Vydavateľské údaje: Amsterdam Elsevier B.V 01.04.2022
Elsevier BV
Elsevier
Predmet:
Computer architecture
Convolutional autoencoder
ENGINEERING
Graph convolution
Modelling
Neural networks
Nonlinear dimensionality reduction
Parametric PDEs
Reduced order models
Reduced-order modeling
System effectiveness
Reduced-order modeling
Parametric PDEs
65M60
Graph convolution
65M22
Convolutional autoencoder
Nonlinear dimensionality reduction
ISSN:0045-7825, 1879-2138
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:The popularity of deep convolutional autoencoders (CAEs) has engendered new and effective reduced-order models (ROMs) for the simulation of large-scale dynamical systems. Despite this, it is still unknown whether deep CAEs provide superior performance over established linear techniques or other network-based methods in all modeling scenarios. To elucidate this, the effect of autoencoder architecture on its associated ROM is studied through the comparison of deep CAEs against two alternatives: a simple fully connected autoencoder, and a novel graph convolutional autoencoder. Through benchmark experiments, it is shown that the superior autoencoder architecture for a given ROM application is highly dependent on the size of the latent space and the structure of the snapshot data, with the proposed architecture demonstrating benefits on data with irregular connectivity when the latent space is sufficiently large. •Autoencoder neural networks facilitate effective nonlinear methods for reduced-order modeling (ROM).•Many popular ROMs employ a standard convolutional autoencoder which is only suitable for regular data.•A new ROM architecture is proposed which uses a graph convolutional autoencoder suitable for irregular data.•The proposed architecture is compared to ROMs based on POD, standard convolutional, and fully connected architectures.•Experiments show that the notion of best architecture is task-dependent, with the proposed ROM producing superior results in several cases.
Bibliografia:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
SC0020270; SC0020418
USDOE Office of Science (SC), Advanced ScientificComputing Research (ASCR). Scientific Discovery through Advanced Computing (SciDAC)
ISSN:0045-7825
1879-2138
DOI:10.1016/j.cma.2022.114764