The accurate estimation of the State of Health (SOH) of lithium-ion batteries is essential for ensuring their safe and reliable operation, as direct measurement is not feasible. This paper presents a novel SOH estimation method that integrates Particle Swarm Optimization (PSO) with an Extreme Learning Machine (ELM) to improve prediction accuracy. Health Indicators (HIs) are first extracted from the battery’s charging curve, and correlation analysis is conducted on seven indirect HIs using Pearson and Spearman coefficients. To reduce dimensionality and eliminate redundancy, Principal Component Analysis (PCA) is applied, with the principal component contributing over 94% used as a fusion HI to represent battery capacity degradation. PSO is then employed to optimize the weights (ε) between the input and hidden layers, as well as the hidden layer bias (u) in the ELM, treating these parameters as particles in the PSO framework. This optimization enhances the ELM’s performance, addressing instability issues in the standard algorithm. The proposed PSO-ELM model demonstrates superior accuracy in SOH prediction compared with ELM and other methods. Experimental results show that the mean absolute error (MAE) is 0.0034, the mean absolute percentage error (MAPE) is 0.467%, and the root mean square error (RMSE) is 0.0043, providing a valuable reference for battery safety and reliability assessments.