Abstract:

To overcome the limitations of traditional state of health (SOH) estimation methods—such as inadequate feature extraction, nonlinear complexity, and difficulty in model parameter optimization—this study proposes a novel SOH estimation approach based on incremental capacity analysis combined with variational mode decomposition (VMD), grey wolf optimization (GWO), and kernel extreme learning machine (KELM). First, an improved voltage-capacity model based on the Lorentz function is employed to fit voltage-capacity data during the constant-current charging process, enabling the extraction of health indicators such as peak voltage, peak value, and peak area. Model parameters are optimized using the GWO algorithm, thereby improving feature extraction accuracy and robustness. Next, VMD is applied to decompose SOH-related signals into multiple intrinsic mode functions. These components serve as inputs to individual sub-models, effectively capturing signal characteristics across distinct frequency domains while mitigating noise and mode mixing. Subsequently, the GWO algorithm is used to optimize the key parameters of the KELM model, significantly enhancing its nonlinear regression capability and estimation accuracy. The proposed method is evaluated through comparative analyses across different training data sizes, estimation models, and datasets from multiple batteries. Experimental results demonstrate that the proposed method achieves high-accuracy SOH estimation using only 100 cycles of data, with a mean absolute error of 0.9751% and a maximum error of 1.9340%. The model also exhibits strong robustness and generalization performance.

Key words: lithium-ion batteries, state of health, incremental capacity analysis, variational mode decomposition, grey wolf optimization, kernel extreme learning machine