An LSTM-Based Approach For Capacity Estimation on Lithium-ion Battery

With the wide application of lithium-ion batteries, the safety and stability of batteries have attracted much attention in recent years. Accurate capacity estimation can help understand the health status of lithium-ion batteries. Model-based methods can perform well on capacity estimation with enough prior knowledge. However, the internal electrochemical mechanism of lithium-ion battery degradation is so complicated that the cost of prior knowledge is usually expensive. Due to no need for prior knowledge, data-driven methods introduce new solutions to solve this problem. Among these, deep learning methods (DL) show great potential for their advantages of high-dimensional feature mapping. In this paper, an LSTM (Long Short-Term Memory) network model is proposed to estimate the capacity, in which seven time-related health features in the charge stage are used as input of the model. On the one hand, health features extracted from charge time data can decrease the difficulties in data acquisition. On the other hand, the LSTM model has the ability to learn the long-term dependencies between data so that suit for constructing mapping relationship between health features and capacity. The validation experiments are conducted based on two simulation datasets. The results show that the proposed model can accurately track the capacity recovery effect and estimate the RUL of lithium-ion batteries.