A novel feature extraction approach for battery health assessment

State of health (SOH) estimation represents a fundamental challenge in lithium-ion battery management systems. Current data-driven approaches predominantly require complete voltage-capacity profiles within predetermined voltage intervals, creating significant barriers for real-world deployment where charging and discharging patterns exhibit inherent randomness. This work develops a feature extraction technique capable of processing partial voltage-capacity data for reliable SOH assessment. The approach constructs capacity increment matrices from available partial curves and employs principal component analysis to derive robust aging indicators. Support Vector Regression validates the extracted features’ effectiveness using NASA battery datasets comprising two groups tested under different thermal conditions (room temperature and 43°C) with varying discharge protocols. Results demonstrate that the proposed features can achieve SOH prediction accuracy of 0.0445 RMSE and 4.67% MAPE across diverse operational conditions, even when utilizing incomplete operational data.