Data-driven state-of-health estimation for lithium-ion battery based on aging features

Reliable state-of-health (SOH) estimation is crucial to the safe operation of lithium-ion battery. Data-driven SOH estimation becomes a hot research topic with the booming of high-performance machine learning algorithms. The effectiveness of a data-driven approach will be enhanced significantly if the input features are extracted properly. In order to improve the SOH estimation accuracy, the features highly associated with battery degradation should be utilized in the data-driven model. In this paper, an aging feature extraction method based on electrochemical model (EM) is proposed to account for the battery degradation mechanisms. The aging features of EM such as charge transfer resistance, solid phase diffusion coefficient and electrode volume fraction are defined as internal health features (IHFs) for SOH estimation. Moreover, external health features (EHFs) are directly extracted from the voltage and temperature curves that split into multiple stages. Then, IHFs and multi-stage EHFs are selected appropriately for SOH estimation in offline and online application scenarios. Finally, two well-known machine learning algorithms are employed to construct data-driven SOH estimation model using IHFs and EHFs. Experimental data are used to prove that the proposed method can effectively improve the accuracy of SOH estimation under different application scenarios and battery charge-discharge modes.

Automated summary

This paper proposes an aging feature extraction method based on an electrochemical model to account for battery degradation mechanisms. Internal health features (IHFs) such as charge transfer resistance, solid phase diffusion coefficient, and electrode volume fraction are defined, along with externally extracted health features (EHFs) from voltage and temperature curves. These features are used in data-driven SOH estimation models constructed using two machine learning algorithms, and experimental data proves the method's effectiveness in improving estimation accuracy under different scenarios and charge-discharge modes.