Early Prediction of Lithium‑Ion Battery Degradation Using Health Indicators and Random Forest Regression

This study investigates early prediction of lithium‑ion battery degradation using health indicators and Random Forest regression, focusing on Tyfast’s lithium vanadium oxide (LVO) cells designed for extreme‑rate charging. LVO batteries were cycled at 0.5C and 10C to evaluate electrochemical performance, thermal behavior, and long‑term stability. Voltage‑capacity curves and differential features (dV/dQ, dQ/dV) were extracted to capture subtle degradation signatures. After data cleaning and normalization, a Random Forest Regressor was trained on cycle count, capacity retention, and voltage‑derived features. The model achieved high accuracy, with R² = 0.998, MAE = 0.72%, and RMSE = 0.90%. Feature importance analysis identified capacity retention and cycle number as the strongest predictors of remaining useful life (RUL), estimating an RUL of ~6,883 cycles at 87.5% retention. Compared with NASA’s 18650 Li‑ion dataset, LVO cells demonstrated significantly slower degradation and superior fast‑charging stability, maintaining ~93% capacity after 1,000 cycles at 10C—far outperforming commercial NMC and LFP cells. These results highlight the promise of data‑driven prognostics for battery health monitoring and the exceptional durability of LVO chemistry.