Power batteries state of health estimation of pure electric vehicles for charging process

Abstract Under different usage scenarios of various electric vehicles (EVs), it becomes difficult to estimate the battery state of health (SOH) quickly and accurately. This paper proposes a SOH estimation method based on EVs' charging process history data. First, data processing processes for practical application scenarios are established. Then the health indicators (HIs) that directly or indirectly reflect the driver's charging behavior in the charging process are used as the model's input, and the ensemble empirical mode decomposition (EEMD) is introduced to remove the noise brought by capacity regeneration. Subsequently, the maximum information coefficient (MIC) - principal component analysis (PCA) algorithm is employed to extract significant HIs. Eventually, the global optimal nonlinear degradation relationship between HIs and capacity is learned based on Bayesian optimization (BO)-Gaussian process regression (GPR). Approximate battery degradation models for practical application scenarios are obtained. This paper validates the proposed method from three perspectives: models, vehicles, and regions. The results show that the method has better prediction accuracy and generalization capability and lower computational cost, which provides a solution for future online health state prediction based on a large amount of real-time operational data.