Estimation of State-of-Health for Lithium-Ion Batteries Using Relaxation Voltage against Variable Operating Conditions

The health assessment of lithium-ion batteries is crucial for the efficient operation of electric vehicles and energy storage systems. However, in the complex and ever-changing operating environment, accurately assessing the health state of lithium-ion batteries remains a challenge that urgently needs to be overcome. This paper proposes a SOH estimation method based on the relaxation phase curves after battery charging. It extracts health features from the relaxation phase and combines them with Gaussian process regression. The method initially extracts six statistical features from the relaxation voltage and selects those with high correlation as inputs for the model. Next, it constructs new combined kernel functions by randomly pairing 5 commonly used kernel functions. It employs cross-validation to adaptively select the optimal combined kernel function. The proposed method is validated with 55 batteries. Results show that compared with traditional single-kernel models, the mean absolute error of the test set is reduced by 0.0243 Ah. Comparative studies with four classic machine learning methods confirm that this method achieves higher evaluation accuracy, verifying its effectiveness.