State-of-Health Estimation for Lithium-Ion Batteries Based on Initial Voltage Segmentation and Transfer Learning

• • The improved TrAdaBoost.R2 algorithm achieves an RMSE of 0.009 in SOH estimation, a 59% reduction compared to classical TrAdaBoost (0.022) and a 50% reduction versus Transfer Stacking (0.018). This precision is critical for grid-scale battery management systems where SOH errors below 1% directly impact dispatch decisions and warranty compliance. • • Feature extraction relies on IC curve values at fixed voltage points, with Pearson correlation analysis determining the optimal voltage difference ΔU. This segmentation approach enables SOH estimation from partial charging data (SOC 20%–80%), eliminating the need for full charge-discharge cycles that are impractical in operational storage plants. • • The integration of DTW-derived similarity into the TrAdaBoost.R2 weight update mechanism accelerates convergence and reduces computational overhead. This is industrially significant for real-time battery monitoring, where latency above 100 ms can degrade control loop performance. • • Validation on NASA B0005 and XJTU Batch-1 datasets confirms the method's generalizability across different cell chemistries and aging trajectories. The RMSE of 0.009 translates to a 0.9% SOH estimation error, which is within the 2% tolerance required for second-life battery grading and repurposing.