Skip to content

Remaining Useful Life (RUL) estimation of Lithium-ion batteries using deep LSTMs

License

Notifications You must be signed in to change notification settings

MichaelBosello/battery-rul-estimation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

49 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Battery-state-estimation

Estimation of the Remaining Useful Life (RUL) of Lithium-ion batteries using Autoencoders + LSTMs and Autoencoders + CNNs.

Introduction

This repository provides the implementation of deep networks for RUL estimation. The experiments have been performed on two datasets: the NASA Randomized Battery Usage Data Set and the UNIBO Powertools Dataset.

Paper (Energies publication)

If you use this repo, please cite our paper:

To Charge or to Sell? EV Pack Useful Life Estimation via LSTMs, CNNs, and Autoencoders [URL]

@Article{en16062837,
    AUTHOR = {Bosello, Michael and Falcomer, Carlo and Rossi, Claudio and Pau, Giovanni},
    TITLE = {To Charge or to Sell? EV Pack Useful Life Estimation via LSTMs, CNNs, and Autoencoders},
    JOURNAL = {Energies},
    VOLUME = {16},
    YEAR = {2023},
    NUMBER = {6},
    ARTICLE-NUMBER = {2837},
    URL = {https://www.mdpi.com/1996-1073/16/6/2837},
    ISSN = {1996-1073},
    DOI = {10.3390/en16062837}
}

Source code structure

The package data_processing contains the scripts that load the data from the two sets. unibo_powertools_data.py loads the data from the UNIBO dataset and compute the derived columns like the SOC one, while model_data_handler.py prepare the time series. nasa_random_data.py both loads and prepares the data of the NASA Randomized set. prepare_rul_data.py is used for both datasets; it calculates the integral of the current to obtain the RUL based on Ah, and it format the time series for the neural network.

The experiments directory contains the Jupyter notebooks defining the various experiments and models used. The results directory shows the plots of the results and the measurements like RMSE, MAE, etc.

The trained models are available in the GitHub release section.

Quick start

1) Install requirements

Python packages

Tested with Tensorflow 2.7

pip install tensorflow
pip install pandas sklearn scipy
pip install plotly
pip install jupyter notebook ipykernel jupyterlab

2) Download the datasets

Download the NASA Randomized Battery Dataset and put its content in the directory battery-state-estimation/data/nasa-randomized/

Download the UNIBO dataset and put its content in the directory battery-state-estimation/data/unibo-powertools-dataset/

3) Run one of the notebooks

Run one of the notebooks in the experiments directory. You can switch between training a new model or loading an existing one by toggling the value of IS_TRAINING at the top of the notebook (just define the model file name in RESULT_NAME).

Check out the results directory if you want to see the results obtained by us (you can find the trained models in the release section of GitHub, you can run them by putting the files into the 'trained-model' directory).

If you want to run the notebook on Google Colab, load the repository in your Google Drive and set to True the variable IS_COLAB at the top of the notebook. This will allow the notebook to find the datasets and to save the results in your Drive.