State of Charge Estimation for Onboard Lithium-ion Batteries Based on an Innovative Network

The low accuracy and weak robustness of SOC estimation of on-board lithium-ion batteries lead to frequent problems such as driver's worry about driving range, shortened battery life, and fire and explosion of electric vehicles. However, the internal electrochemical reaction of lithium-ion batteries is complex, and the external application scenarios are random and changeable, making SOC estimation challenging. This paper proposes a SOC estimation method for onboard lithium-ion batteries based on the CNN-GRU-Attention hybrid network. Firstly, the real vehicle data is preprocessed to obtain the available discharge segments. Then, three simple features with strong correlation with SOC are selected as the input of the hybrid network by using the Pearson-Spearman feature double criteria analysis. Finally, the hybrid network is constructed and tested on the real vehicle operation data in the northern and southern regions respectively. The experimental results show that the average absolute error and root mean square error of SOC estimation are both less than 1.6%, which verifies the feasibility and effectiveness of this method.