PINN4SOH

This code is for our paper: Physics-informed neural network for lithium-ion battery degradation stable modeling and prognosis

Important

Summary of articles using the XJTU Battery Dataset: https://github.com/wang-fujin/XJTU-Battery-Dataset-Papers-Summary

1. System requirements

python version: 3.7.10

Package	Version
torch	1.7.1
sklearn	0.24.2
numpy	1.20.3
pandas	1.3.5
matplotlib	3.3.4
scienceplots

2. Installation guide

If you are not familiar with Python and Pytorch framework, you can install Anaconda first and use Anaconda to quickly configure the environment.

2.1 Create environment

conda create -n new_environment python=3.7.10

2.2 Activate environment

conda activate new_environment

2.3 Install dependencies

conda install pytorch=1.7.1
conda install scikit-learn=0.24.2 numpy=1.20.3 pandas=1.3.5 matplotlib=3.3.4
pip install scienceplots      # for beautiful plots

3. Demo

We provide a detailed demo of our code running on the XJTU dataset.

1. Run the main_XJTU.py file to train our model. The program will generate a folder named results and save the results in it.
2. Run the main_comparison.py file. You can change setattr(args,'model','MLP') to select the CNN or MLP model. It will generate a folder in the results to save the results of the corresponding model (CNN or MLP).
3. Run the results analysis/XJTU results.py file. It will process the results in Step one and generate the XJTU_results.xlsx file. At the same time, the results of each batch in the XJTU dataset will also be printed on the Command Console, corresponding to the results in Table 2 of our manuscript.
4. Run the results analysis/Comparision results.py file to generate the XJTU-MLP_results.xlsx file and save it in the results folder. The results of each batch in the XJTU data set will also be printed on the Command Console, corresponding to the results in Table 2 of our manuscript.

Note: As we all know, the training process of neural network models is random, and the volatility of regression models is often greater than that of classification models. Therefore, the results obtained from the above process are not expected to be exactly identical to those mentioned in our manuscript. However, it is evident that the results obtained from our method are superior to those of MLP and CNN.

In addition, we also provide the results of our training, which are saved in the results folder and results analysis folder. These results correspond exactly to the data in our manuscript.

What's more, we also provide the codes corresponding to the Figures in our manuscript, which are saved in the plotter folder. You can use these codes to draw the Figures in the manuscript.

4. Additional information

The data in the data folder is preprocessed data. Raw data can be obtained from the following links:

1. XJTU dataset: link
2. TJU dataset: link
3. HUST dataset: link
4. MIT dataset: link

The code for reading and preprocessing the dataset is publicly available at https://github.com/wang-fujin/Battery-dataset-preprocessing-code-library

---

We generated a comprehensive dataset consisting of 55 lithium-nickel-cobalt-manganese-oxide (NCM) batteries.

It is available at: Link

Zenodo link: https://zenodo.org/records/10963339.

5. Citation

If you find it useful, please cite our paper:

@article{wang2024physics,
  title={Physics-informed neural network for lithium-ion battery degradation stable modeling and prognosis},
  author={Wang, Fujin and Zhai, Zhi and Zhao, Zhibin and Di, Yi and Chen, Xuefeng},
  journal={Nature Communications},
  volume={15},
  number={1},
  pages={4332},
  year={2024},
  publisher={Nature Publishing Group UK London}
}