Survival-Analysis Based Prediction of Remaining Useful Life of Lithium-Ion Batteries with an Analysis of Input Feature Effects

Predicting the Remaining Useful Life (RUL) of lithium-ion batteries is crucial for the reliability of applications like energy storage systems (ESS) and electric vehicles. Conventional studies predominantly rely on regression but often limit the interpretation of degradation as a time-to-event process. This study reformulates RUL prediction using survival analysis, proposing a probabilistic model based on observed degradation states. This approach characterizes time-dependent risk, interpreting lifetime as relative hazard rather than absolute lifespan, which offers greater flexibility in capturing heterogeneous degradation patterns. Furthermore, we systematically analyze the impact of input feature configurations on prediction performance. The proposed approach introduces a new perspective on RUL prediction, applicable to battery health monitoring and preventive maintenance.