Data-driven methods for estimating the state of health (SOH) of lithium-ion batteries (LIBs) are widely used by extracting health indicator (HI) from charge-discharge measurements. However, many existing HIs have shortcoming of heavy computing burden, which causes the difficulty on online implementation. Therefore, this paper proposes a novel HI called Equal Voltage Range Sampling Count Number (EVRSCN), which is used to estimate SOH of LIBs. The proposed HI is extracted from the charging process. The EVRSCN HI can be extracted online with reduced calculation burden of BMS. Gaussian Process Regression (GPR) is used to quickly achieve accurate SOH estimation based on EVRSCN. The SOH estimation which applies three typical and widely-used datasets of Oxford, SNL and CALCE shows that the proposed method can achieve a promised accuracy, and the root-mean-square error (RMSE) could lower than 0.5% in some typical cases. In addition, the noise immunity of EVRSCN has been evaluated and compared with existing HIs. The results show that the proposed EVRSCN has stable and promised SOH estimation accuracy, as well as good noise immunity.
11 December 2024 in IEEE Transactions on Transportation Electrification
Estimating the state of health (SOH) of lithium-ion batteries (LIBs) based on data-driven methods are widely used by extracting health features (HFs) from complete charging measurements. Since the user’s charging habits are different, it is, however, difficult to obtain complete HFs under random charging conditions. This article proposes an SOH online estimation method based on random charging curves to solve this problem. First, rough localization of the random charging curve is accomplished to obtain its position relative to the full charging curve and get the data needed for the reconstruction model by backtracking the historical charging data. Second, the extreme learning machine optimized by genetic algorithm and hill climbing algorithm (GA-HC-ELM) is applied to reconstruct the partial charging curve. Then, HFs are extracted from the reconstructed curves, and Gaussian process regression (GPR) is applied for fast SOH estimation. Finally, three commonly used public datasets were selected to conduct a feasibility analysis of the proposed method. The results show that the charging curve reconstruction average root-mean-square error (RMSE) is within 1.01 mV in some typical cases, and the SOH estimation RMSE is stable within 1.22% in most cases, indicating that this method has a reliable SOH estimation effect under random charging conditions.