A Data-Driven GRU-GPR Method for Predicting the Remaining Useful Life of Lithium-Ion Battery

Remaining useful life prediction is crucial in the prognostics and health management of lithium-ion batteries. This paper combines the gated recurrent unit with Gaussian process regression to predict the remaining useful life of batteries based on limited training data. The proposed method can capture the non-linear degradation of capacity and provide corresponding uncertainty estimation. The approach begins with relevant feature selection by least absolute shrinkage and selection operator. Subsequently, these selected features undergo empirical model decomposition for denoising, resulting in intrinsic mode functions that serve as inputs to the predictive model. The validation experiments are conducted based on batteries from the data repository of NASA. Comparative results demonstrate that the developed model achieves low RUL prediction errors of 0.4, outperforming other benchmark methods. Furthermore, the pro-posed method exhibits robustness in predicting complex declining capacity patterns, even with limited training data.