A power model considering initial battery state for remaining useful life prediction of lithium-ion batteries

Square-root-of-time model, constructed based on the growth of solid electrolyte interface layer, is an extensively-used semi-empirical model for remaining useful life (RUL) prediction of lithium-ion batteries. However, over the life cycle, the battery capacity degradation is not always under a linear relationship to the 1/2 power of the cycle number. In practice, its initial state, fresh or old, is rarely considered during RUL prediction. To address these concerns, a three-step mathematical transformation is proposed to improve the flexibility of square-root-of-time model. With initial battery state described by an initial cycle parameter, a power model is proposed to capture the battery capacity degradation. The parameter properties of proposed power model are then discussed in depth. Combining an offline parameter estimator and an online particle filter algorithm, a two-phase prediction framework is developed for onboard RUL prediction. Finally, a charge-discharge experiment is conducted, and its comprehensive experimental datasets of lithium iron phosphate batteries are analyzed. Results show that the proposed power model is superior to other existing degradation models on model fitting and extrapolation accuracy; and compared to the traditional square-root-of-time model, the RUL prediction accuracy is significantly improved.