Fast capacity prediction of lithium-ion batteries using aging mechanism-informed bidirectional long short-term memory network.

• A physical informed neural network is built for aging mechanism identification. • A hybrid prediction method combining the aging mechanism and data-driven method. • Entire charging curve prediction using developed deep hybrid neural network. • Fast capacity prediction only using 50 mV charging voltage window. Accurate and robust capacity prediction is significant for battery management system to identify the state of health and life condition for lithium-ion batteries. This paper proposes a fast capacity prediction method by developing a novel deep aging mechanism-informed bidirectional long-short term memory (AM-Bi-LSTM) neural network. Firstly, a physical informed aging mechanism (AM) layer is established with the random charging curve sequences as input to identify the degradation features. Then the deep learning framework with two bidirectional long-short term memory (Bi-LSTM) layers is built to reflect the entire constant current charging curves and predict the battery capacity. In which, the battery aging mechanism is integrated into the artificial intelligence algorithm of capacity prediction for the first time. Several case studies are implemented to verify the effectiveness of developed method, and the influence of voltage window length on capacity prediction is further discussed. The results demonstrate that the charging curves can be accurately and fast captured with a capacity prediction root mean square error of less than 0.49% for 0.74 Ah batteries with 50 mV voltage window charging points collected in only less than 2.09 minutes mean cost time in the whole life cycle. It shows the proposed aging mechanism-informed data-driven prediction method has stronger robustness, faster prediction speed and higher accuracy compared with other data-driven methods. [ABSTRACT FROM AUTHOR]