A stage-based adaptive activation function model for the state of health prediction of vanadium redox flow batteries

This is the authors' abstract. We don't add key points for this paper.

In flow battery systems, the battery management system (BMS) plays a crucial role in ensuring the safety, stability, and efficient operation of the overall system. Among the key indicators monitored by the BMS, the state of health (SOH) represents the current health condition of the battery. Therefore, accurate estimation of SOH is essential for reflecting the degradation level of the battery and predicting its remaining useful life. In this study, the variation in the grey relational degree between input features and SOH is analyzed to identify the stage transition points of SOH degradation. Based on this analysis, a three-stage SOH prediction framework is established using the extreme learning machine (ELM) as the core prediction model. Furthermore, the activation function is improved to construct both discrete adaptive activation functions and continuous adaptive activation functions. Through 30 independent experiments for comparative evaluation, the results demonstrate that the model with the discrete adaptive activation function achieves the best predictive performance. The obtained mean absolute error (MAE), root mean square error (RMSE), and maximum absolute error (MaxAE) are 0.14%, 0.26%, and 2.07%, respectively, indicating that the proposed model can achieve high prediction accuracy even under small-sample training conditions.