Skip to content

Latest commit

 

History

History
69 lines (55 loc) · 3.26 KB

README.md

File metadata and controls

69 lines (55 loc) · 3.26 KB

Synthetic Battery Data

Generation of synthetic battery operation data using a generative adversarial network (GAN) and a state-of-charge (SOC) estimator.

Introduction

This repository provides the implementation of a fusion approach combining a GAN and an SOC estimator for synthetic battery operation data generation. The experiment is performed on the public battery dataset: LG 18650HG2 Li-ion Battery Data.

Publication

The work shown here is an implementation from a research article that was published in the Electronics. If you use this work, please cite our paper:

A Novel Fusion Approach Consisting of GAN and State-of-Charge Estimator for Synthetic Battery Operation Data Generation [DOI: 10.3390/electronics12030657]

@Article{electronics12030657,
AUTHOR = {Wong, Kei Long and Chou, Ka Seng and Tse, Rita and Tang, Su-Kit and Pau, Giovanni},
TITLE = {A Novel Fusion Approach Consisting of GAN and State-of-Charge Estimator for Synthetic Battery Operation Data Generation},
JOURNAL = {Electronics},
VOLUME = {12},
YEAR = {2023},
NUMBER = {3},
ARTICLE-NUMBER = {657},
URL = {https://www.mdpi.com/2079-9292/12/3/657},
ISSN = {2079-9292},
DOI = {10.3390/electronics12030657}
}

Source code structure

  • data directory: contains the preprocessed data lg_600_data.npy of LG 18650HG2 Li-ion Battery Data that is reshape into the form of $[n,600,4]$. Please refer to the SOC estimator repository for detail handling of the preprocessing.
  • result directory: the training result will be output to this directory.
  • soc_models directory: contains the pre-trained model of the SOC estimator used in this work (download from here).
  • arg_parser.py: code to parse command line arguments.
  • lg_dataset.py: code to handle the LG dataset.
  • gan_trainer_base.py: code to handle the basic GAN training procedure.
  • train_lstm_gan.ipynb: jupyter notebook for evaluating the training result.
  • train_lstm_gan.py: main code for the proposed GAN model.

Quick start

Install requirements

Create conda environment from our yml file. Depending on your server's configuration, the version of PyTorch and TensorFlow may differ.

conda env create -f environment.yml

Available arguments

Use below command to view the available arguments. The default arguments are the parameters used in our paper.

python3 train_lstm_gan.py -h

Training

Start training with the default parameters.

python3 train_lstm_gan.py

Evaluation

Once you start the training, a new result will be created in the results directory. Run the result.ipynb notebook to view the training result. Alternatively, you can download our result form release and place to the results directory to evaluate our trained model that was presented in the paper.