A hybrid battery degradation model combining arrhenius equation and neural network for capacity prediction under time-varying operating conditions

Capacity degradation modeling and remaining useful life (RUL) prediction play a pivotal role in enhancing the safety of battery energy storage systems. Lithium-ion batteries utilized in the systems are subjected to intricate and variable operating conditions, including temperature, charging/discharging rates and depth, which result in time-varying degradation rates. However, most existing models for battery capacity prediction have the limitation of ignoring the quantitative effects of time-varying operating conditions on degradation. Therefore, a hybrid battery capacity degradation model combining Arrhenius equation and lightweight Transformer is constructed. The effects of operating conditions on capacity degradation rates are introduced by an improved Arrhenius degradation equation. The coordinate ascent method embedded with Newton descent is used for estimating the model parameters. Then, the unknown mapping relationships between the degradation statistical features extracted from monitoring data and the degradation factors are captured by the light-weight Transformer. Finally, a novel framework consisting of future capacity and RUL predicting is constructed based on sequential importance resampling (SIR). For the purpose of verification, the cyclic charge-discharge experiments of eight battery cells under two distinct operating profiles are designed and conducted. Compared with other models, the proposed model achieves the best prognostics performance on the tested cells.