A data-driven method for state of health prediction of lithium-ion batteries in a unified framework

Abstract

State of health (SOH) is paramount performance for the management and maintenance of lithium-ion batteries. SOH prediction is of great significance to state of charge (SOC) estimation and remaining useful life (RUL) prediction. To achieve accurate SOH prediction in a unified framework including one-step, multi-step and long-term prediction, empirical mode decomposition (EMD) is introduced to diminish the local fluctuation, and then the decoupled residual SOH series is regarded as the training series. The optimized dynamic single-exponential model is used to describe the SOH degradation. Subsequently, the optimal system state is determined by particle filter (PF) algorithm. Effectiveness and dominant of this method are validated via the multiple simulation comparative experiments based on National Aeronautics and Space Administration (NASA) data set. Additionally, one-step, multi-step and long-term SOH prediction performance are analyzed in detail. The results indicate that the proposed method realizes a unified prediction framework with uncertainty representation for lithium-ion batteries and outperforms other important methods with higher prediction accuracy. Furthermore, the prediction results are still considerable even with a small number of historical SOH data. Note that this method is also employed in other similar battery management systems.