Implementation of the convolutional autoencoder echo state netowrk (CAE-ESN) applied to the Kolmogorov flow, based on the paper by A.Racca et al. https://arxiv.org/abs/2211.11379.
The CAE-ESN is a methodology to predict spatiotemporal dynamics, such as two dimensional flows, from data. The convolutional autoencoder provides a mapping of the high-dimensional flowfield to a low-order representation, the latent space, whose dynamics are predicted by the ESN.
-
In tutorial, a simplified version of the architecture, which can be trained and tested in ~20 mins, is explained in detail through Jupyter notebooks.
-
In source, you can find the scripts used to train the architeture in the paper.
More details are provided by the Readme file inside the tutorial and source folders.
The code requires Tensorflow 2.6.