Lithium-Ion Battery Remaining Useful Life Prediction Based on GRU-RNN

Lithium-ion battery has been widely applied as an energy storage component in various industrial applications including electric vehicles, distributed grids and space crafts. However, the battery performance degrades gradually due to the SEI growth, li-plating and other irreversible electro-chemical reactions. These inevitable reactions directly influence the reliability of the energy storage system and may further cause catastrophic consequences to the host system. Remaining useful life (RUL) is one of critical indicators to evaluate the battery performance. This paper proposes a battery RUL prediction approach based on a new recurrent neural network (RNN), i.e. the RNN with Gated Recurrent Unit (GRU). The proposed method overcomes the drawback on dealing with long term relationship of RNN. The structure of the RNN-GRU is much simpler which contributes to a higher computational complexity. The experiments based on the NMC lithium-ion battery cycle life testing data are conducted and the results indicate that the mean error of different battery cells are both less than 3% which means the proposed method is accurate and robust for battery RUL predictions.