Abstract: Accurate prediction of the remaining useful life (RUL) of lithium-ion batteries is crucial for automotive battery management systems. Addressing the issue of inaccurate RUL prediction for lithium-ion batteries, firstly, we extract multiple health factors (HF) using Incremental Capacity (IC) curves, which are sensitive to battery degradation trends, at various constant current charging voltage intervals. We then analyze the correlation between these multiple HFs and battery capacity using the Spearman coefficient method. Secondly, to address the issue of the Extreme Learning Machine (ELM) parameters easily falling into local optima, leading to weak model predictive performance stability, we propose the Subtraction-Average-Based Optimizer (SABO) algorithm to optimize the weights and bias thresholds of the ELM model, thereby improving the model's predictive performance. Finally, we validate the proposed method using publicly available battery datasets from NASA. Experimental results demonstrate that under multiple health factors, employing SABO to optimize ELM improves RUL prediction accuracy, with prediction errors controlled within 2%.