Comparison of Remaining Life Prediction of LiFePO4 Batteries Using Machine Learning Based on Capacity Degradation Analysis

In the electrification era, batteries are vital in innovating sustainable renewable energy utilization. Still, the current challenge is to present a durable, safe battery that does not degrade quickly, does not break soon, and can indicate its age. Additionally, proper charging is crucial for optimizing battery performance. LiFePO4 batteries are chosen because they are safe, environmentally friendly, and have a good remaining life. The Machine Learning (ML) model, combined with the Thevenin model, is optimized in this study to analyze battery cell degradation and predict the Remaining Useful Life (RUL). This study utilizes battery capacity data under new conditions, specifically 90 cycles for four cells and 100 cycles for one cell. Three indicators - percentage of degradation, RUL, and R² (the coefficient of determination) - are selected to assess the accuracy of the prediction results. This study found that the amount of degradation and RUL value depend on the number of cycles and the performance of the ML model applied. Linear Regression 90 cycles on Battery 1 and Random Forest model 100 cycles on Battery 5 have successfully proven that the model is most suitable in predicting the remaining life from capacity degradation with a model evaluation value of 98.28% with RUL 16271 cycles on the Random Forest model and degradation of 0.065% with RUL 27786 cycles on the Linear Regression model.