Code of the paper: "A predictive hybrid reduced order model based on proper orthogonal decomposition combined with deep learning architectures"
This repository contains the code and dataset files for the paper: "A predictive hybrid reduced order model based on proper orthogonal decomposition combined with deep learning architectures", R. Abadia-Heredia, M. Lopez-Martin, B. Carro, J.I. Arribas, J.M. Perez and S. Le Clainche.
All code files are available in this (github) repository. The description of the different files is as follows:
- Cilin3D_20 v2.1.ipynb: It provides an implementation of the CNN1D and LSTM models for the cilind3d_Mat dataset. It includes the initial SVD transformation. The code provides the results obtained with the original full dataset. If the code is executed with the reduced version of the dataset that is provided here (link below), there are a few lines of code that need to be changed. These lines of code are explicitly marked.
The code is prepared to make predicitions with k snapshots used as predictors to predict p time-ahead snaphots. In particular, in the code are used: p = 6 and k = 10. The results included in the paper present only the first time-ahead prediction (of the p available).
The code and the dataset provided in this repository represent a representative example of the models and datasets used in the paper. A reduced version of the original dataset used in the paper is provided in the following repository: https://data.mendeley.com/datasets/stdf3z6p2f/2
The code can be executed in the cloud using Google Colaboratory https://colab.research.google.com/notebooks/intro.ipynb