Data and code for the paper "Ultra-early prediction of lithium-ion battery performance using mechanism and data-driven fusion model"
This version in python is available on https://github.com/swhlqu/Ultra-early-performance-prediction.
The simulated dataset using COMSOL is available on COMSOL_DATASET. It consists the simulated data of NCM523 under charging rates of 0.5, 0.6, 0.7, 0.8, 1.0, 1.5, 2.0 C at 25 ℃, 0.5 C at 35 ℃, 0.6, 0.7, and 1.0 C at 45 ℃. It also includes the simulated data of the NCM811 battery under charging rates of 0.5, 0.7, 0.8, 0.9, 1.0, 1.5, 2.0 C at 25 ℃, 0.7, 0.8, 0.9, and 1.0 C at 35 ℃, and 0.8, 0.9, and 1.0 C at 45 ℃.
Note: The P2D model using COMSOL is available from the corresponding authors upon request.
The actual charging curves over the full life cycle of the NCM523 and NCM811 batteries is available on NCM523 and NCM811 folds, which 25_0.5CC_0.5CD.xlsx refers the charging data is testing under 25 ℃ with 0.5 C charging and discharging rates.
The actual ISC test dataset of the NCM523 and NCM811 battery is available on NCM523-ISC and NCM811-ISC folds, which BD refers to the normal test without paralleled ISC resistance, and CS refers to the ISC test with paralleled ISC resistance. The dataste in BD fold is used to correct the simulated results, and that in CS fold is utilized to detect ISC.
DNN and Attention are the propoed general framework which can predict the charging curves over full life cycles of NCM523 and NCM811 batteries in the case of only knowing the charge protocols.
DNN and Attention requires only a standard computer with enough RAM to support the in-memory operations.
COMSOL Multiphysics pytorch The python package is supported for windows. The package has been tested on the following systems:
windows: windows11
DNN and Attention mainly depends on the Python scientific stack. The data processing and the development of all proposed methods are implemented in Python 3.9 with Pytorch 1.11. The computation is executed based on an RTX 3090 graphics processing unit.
torch numpy pandas time warnings os math pathlib
The dataset divided method can be found in the main text of "Ultra-early prediction of lithium-ion battery performance using mechanism and data-driven fusion model" The training, validation, and test set in each scenario can be found in each scenario fold.
The proposed method has been demonstrated in the scenarios of Charging curve of one cycle prediction, Degradation pattern identification under different working conditions, Degradation pattern identification under different working conditions and temperatures, and Micro-ISC detection.
In the scenario of Charging curve of one cycle prediction and Micro-ISC detection, only the proposed DNN is used to achieve the prediction performance.
In the scenario of Degradation pattern identification under different working conditions and Degradation pattern identification under different working conditions and temperatures, the proposed DNN and attention model are used to achieve the prediction performance. The attention model is used first to obtain the predicted results, then the proposed DNN is utilized to achieve accurate prediction performance.
The best model parameters trained by us are shown in the checkpoint folder in each scenario.
The prediction results can also be found in these folds. The "test_preds.csv" refers to the predicted results of the proposed DNN in the scenario of Charging curve of one cycle prediction and Micro-ISC detection. In the scenario of Degradation pattern identification under different working conditions and Degradation pattern identification under different working conditions and temperatures, the "test_preds.csv" refers to the predicted results of the attention model, and the "test_preds_second.csv" is the predicted results of the proposed DNN.
In the scenario of Micro-ISC detection, the ISC folder includes the predicted results, named "pred_XXCC_XXCD.xlsx", and the actual ISC test dataset paralleled ISC resistance, named "51_point_ISC_523 or 811_CS_XXCC_XXCD_XXohm.csv". Through the error of the ISC test dataset and predicted dataset, the ISC can be detected
This work is a free, open-source software package (distributed under the MIT license).
If you have any questions, you can reach the author at the following e-mail: cydu@hit.edu.cn