Battery state of health estimation based on voltage relaxation and an improved online sequential extreme learning machine

Abstract

Abstract Accurate online estimation of battery state of health (SOH) is crucial for the safety and stability of electronic devices. In reality, the complete operating information is difficult to obtain, making it hard to extract suitable features. Moreover, the online sequential extreme learning machine (OS-ELM) cannot dynamically adjust the model based on the temporal relations of samples, resulting in poor learning ability for temporal features. To address these problems, this paper proposes an OS-ELM with a forgetting learning mechanism (FLOS-ELM). First, features are extracted from the relaxation curves independent of the charging/discharging process. Then, a forgetting learning mechanism is introduced in the OS-ELM to update the model adaptively and improve the ability to capture temporal features and learn online. Finally, a mapping model of degradation features and SOH is constructed via the proposed FLOS-ELM to achieve accurate SOH estimation. Experimental results on the publicly available datasets show that the RMSE is 0.9%, verifying the validity and accuracy of the model.