Co-estimation of state of charge and capacity for lithium-ion battery based on recurrent neural network and support vector machine

Abstract

To accurately estimate the state of charge (SOC) of the aged batteries, the capacity estimation must not be ignored. Based on the battery charging data, this paper proposes a co-estimation model to estimate the SOC and capacity for lithium-ion batteries. First, a new health indicator of capacity is extracted based on the charging data of lithium batteries; second, the capacity is estimated by least squares support vector machine (LSSVM). The results are recorded based on a memory gate and used as the input of SOC estimation. Third, a moving window method is adopted to address the long-term dependency loss problem of the recurrent neural network (RNN), and a co-estimation model is obtained. Finally, the proposed model is compared with other models. The results show that the proposed model performs better. The maximum RMSE of the proposed model is 0.85%, and the computational cost of the proposed model is below 5 s.

Keywords

• Lithium-ion batteries
• State of charge
• Capacity estimation
• Moving window
• Recurrent neural network
• Support vector machine