Kinetic-constrained neural ODE (KCNODE)

This repository supplements our Chemical Engineering Journal paper (doi:10.1016/j.cej.2023.146869) and the earlier ChemRXIV preprint (doi:10.26434/chemrxiv-2023-x39xt).

Contents of the repository

• data_generator.py - Functions for generating the data for the numerical experiment
• KCNODE.py - Classes and methods for building neural ODE models

---

• Baseline model.ipynb - Example using the baseline model
• KCNODE FT.ipynb.ipynb - Example using the KCNODE model for the process of CO₂ hydrogenation to hydrocarbons via FT (real data)
• KCNODE methanation.ipynb - Example using the KCNODE model for the process of CO₂ hydrogenation to methane (numerical experiment)
• Training.ipynb - Example demonstrating training of the neural ODE models

---

• /trained_models - Directory with a trained model of neural ODE
  • baseline.pt - The baseline neural ODE model.
  • KCNODE_methanation.pt - The KCNODE model for CO₂ hydrogenation to CH₄
  • KCNODE_FT.pt - The KCNODE model for CO₂ hydrogenation to hydrocarbons via FT

Used environment

The code was developed on Windows 10 but it should be platform independent.

• Python version: 3.9.12 (amd64)
• Packages:
  • torch:1.12.0
  • torchdyn:1.0.3
  • numpy:1.23.0
  • scipy:1.8.1
  • pandas:1.4.3
  • matplotlib:3.5.2
  • tqdm:4.64.0
  • doepy:0.0.1
  • notebook:6.5.3

Install the dependencies with pip install -r requirements.txt

Authors

• Aleksandr Fedorov (on GitHub fizorg103)
• David Linke (on GitHub dalito)

Acknowledgement

Financial support from German Federal Ministry of Education and Research (BMBF) through the project InnoSyn (FKZ: 03SF0616B) and from German Research Foundation (DFG) through the project "NFDI4Cat - NFDI for Catalysis-Related Sciences" (DFG project no. 441926934) within the National Research Data Infrastructure (NFDI) programme of the Joint Science Conference (GWK) is gratefully acknowledged.