<p>Although machine learning techniques have been widely employed for state-of-health (SOH) estimation of lithium-ion batteries (LIBs) in electric vehicles (EVs), the inherent complexity of LIBs’ reaction mechanisms and their highly nonlinear, time-dependent aging processes present significant challenges. Traditional SOH estimation methods, which depend on complete cycling data, are often impractical in real-world deployment scenarios due to data limitations. To address these issues, we propose a voltage-interval optimized SOH estimation approach based on incremental capacity analysis (ICA) and correlation feature selection. In this method, the charging voltage profile is divided into multiple diagnostic intervals, and the optimal voltage window is determined by analyzing the correlation between incremental capacity curve peak characteristics. This approach not only reduces data redundancy but also ensures high estimation accuracy by selecting degradation-sensitive features. Furthermore, a dynamic neural network-based SOH estimation model is developed and validated using both CALCE and NASA datasets. Experimental results demonstrate that the maximum error in SOH estimation is below 1.5%, and the computation time is reduced by approximately 66.67% compared to conventional methods. The proposed approach shows superior robustness, practical applicability, and strong generalization capability across various battery types and degradation conditions.</p>