TY - JOUR
T1 - A Surrogate Reduced Order Model of the Unsteady Advection Dominant Problems Based on Combination of Deep Autoencoders-LSTM and POD
AU - Kherad , Mahdi
AU - Moayyedi , Mohammad Kazem
AU - Fotouhi , Faranak
JO - Advances in Applied Mathematics and Mechanics
VL - 6
SP - 1410
EP - 1450
PY - 2024
DA - 2024/10
SN - 16
DO - http://doi.org/10.4208/aamm.OA-2022-0163
UR - https://global-sci.org/intro/article_detail/aamm/23473.html
KW - Reduced order model, deep learning, autoencoder, long short-term memory, proper orthogonal decomposition.
AB - <p style="text-align: justify;">Model Order Reduction is an approximation of the main system so that
the simplified system retains important features of the main system. Deep learning technology is a recent breakthrough in artificial neural networks that can find
more hidden information from the data. In this paper, a non-intrusive reduced order model (NIROM) based on combining deep neural networks (DNNs) and POD
abilities, namely FAE-CAE-LSTM is presented. This method combines the obtained
features based on Fully connected autoencoders (FAE), Convolutional autoencoders
(CAE), and POD and then, a deep Long short-term memory network is trained by obtained features to predict the pressure and velocity fields at future time instances. We
investigate the performance of the proposed methodology by solving two well-known
canonical cases: a strong shear flow exhibiting the Kelvin–Helmholtz instability, and
flow past a cylinder. The performance of the proposed FAE-CAE-LSTM method in future state prediction of the flow is compared with other NIROM methods such as CAE-LSTM, autoencoder-LSTM, autoencoder-DMD and POD-RNN based models. Results
show that the FAE-CAE-LSTM method is considerably capable of predicting fluid flow
evolution and obtains the best results in the prediction of the pressure and velocity
fields in future time instances.</p>