| 초록 |
Accurate prediction of battery capacity fade is essential to improve the performance of the battery and ensure the safety of the users. As the lithium iron phosphate(LFP) batteries are put through repeated charging and discharging cyclic processes, the batteries encounter sudden degradation of the State of Health. The point where the sudden degradation starts is called ‘knee-point’. Therefore, early knee-point prediction is essential to know when to replace the battery and prevent battery ignition. In this study, we propose a real-time machine learning models that can predict knee-point 100 cycles before a sudden drop occurs. Since the existing studies use the information of the initial cell cycle have limitations in predicting changes in the battery system when the driver condition changes, it is important to update the relevant VIT information and cycle information in real-time. In this study, two different models, Long short-term memory and Multi-layer perceptron, were proposed and compared. Moreover, the study shows the gap between the 2 models using Kfold validation to show the model influence of remembering information about the state of a battery of the history. |
