huzaifi18 / RUL_prediction

The project focused on "Battery Remaining Useful Life (RUL) Prediction using a Data-Driven Approach with a Hybrid Deep Model combining Convolutional Neural Networks (CNN) and Long-Short Term Memory (LSTM)." This repository aims to revolutionize battery health estimation by leveraging the power of deep learning to predict the remaining useful life

A Hybrid CNN-LSTM for Battery Remaining Useful Life Prediction with Charging Profiles Data

• Battery RUL prediction using data-driven method based on a hybrid deep model of Convolutional Neural Networks (CNN) and Long-Short Term Memory (LSTM).
• CNN and LSTM are used to extract features from multiple measurable data (Voltage, Current, Temperature, Capacity) in parallel.
• CNN extracts features of multi-channel charging profiles, whereas LSTM extracts features of historical capacity data of discharging profiles which related to time dependency.
• This repository provides the code for training in python.

---

Framework:

• Voltage (V), Current (I), and Temperature (T) inputs will each get in the CNN layer separately.
  • Feature V gets into a different CNN layer with features I and T, as well as a feature I get into a separate CNN layer with V and T, and so on.
  • The output from the CNN layer for each feature, is then concatenated. Then they get the next CNN layer
  • The extracted features in the last CNN layer is concatenated with the output of the LSTM layer.

Results

Model	RMSE	MAE	MAPE (%)
SC-LSTM	0,0620	0,0549	3,6080
MC-LSTM	0,0403	0,0340	2,2847
SC-CNN-LSTM	0,0270	0,0215	1,3804
MC-CNN-LSTM	0,0359	0,0291	1,9346
MC-SCNN-LSTM	0,0276	0,0220	1,4207

• SC : Single Channel, MC : Multi Channel
• The performance of prediction models were compared using some evaluation metrics including root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE).
• The hybrid model with excellent feature extraction helps to produce more accurate prediction.
• The MC-SCNNLSTM, MC-CNN-LSTM, and SC-CNN-LSTM model’s prediction results produce predictive values that are close to actual values and are better than the baseline model.
• Hybrid of CNN-LSTM model achieves 61%, 37%, and 15% performance improvements of MAPE in terms of SC-CNN-LSTM, MC-SCNN-LSTM, and MC-CNN-LSTM respectively, compared to using the single model

How to Cite

  @inproceedings{10.1145/3575882.3575903,
  author = {Hafizhahullah, Huzaifi and Yuliani, Asri Rizki and Pardede, Hilman and Ramdan, Ade and Zilvan, Vicky and Krisnandi, Dikdik and Kadar, Jimmy},
  title = {A Hybrid CNN-LSTM for Battery Remaining Useful Life Prediction with Charging Profiles Data},
  year = {2023},
  isbn = {9781450397902},
  publisher = {Association for Computing Machinery},
  url = {https://doi.org/10.1145/3575882.3575903},
  doi = {10.1145/3575882.3575903},
  booktitle = {Proceedings of the 2022 International Conference on Computer, Control, Informatics and Its Applications},
  pages = {106–110},
  numpages = {5},
  keywords = {Lithium-ion battery, remaining useful life, capacity prediction, CNN-LSTM, neural networks},
  }