Energy Storage Science and Technology ›› 2023, Vol. 12 ›› Issue (7): 2229-2237.doi: 10.19799/j.cnki.2095-4239.2023.0292

Previous Articles     Next Articles

SAM-GRU-based fusion neural network for SOC estimation in lithium-ion batteries under a wide range of operating conditions

Hongsheng GUAN(), Cheng QIAN(), Binghui XU, Bo SUN, Yi REN   

  1. School of Reliability and Systems Engineering, Beihang University, Beijing 100191, China
  • Received:2023-04-28 Revised:2023-05-23 Online:2023-07-05 Published:2023-07-25
  • Contact: Cheng QIAN E-mail:guanhs@buaa.edu.cn;cqian@buaa.edu.cn

Abstract:

Accurate estimation of the state of charge (SOC) of lithium-ion batteries under a wide range of operating conditions is crucial for ensuring the operational safety and reliability of electric vehicles; therefore, estimating SOC is one of the most important tasks of battery management systems. In this study, a fusion neural network model combining Self-Attention Mechanism (SAM) and Gated Recurrent Unit (GRU) is proposed to capture the long-term nonlinear mapping relationship between the measurable parameters (voltage and current) and SOC of lithium-ion batteries. This SAM-GRU neural network model makes full use of the short-time processing capability of GRU and the long-time sequence feature-extraction capability of SAM. Additionally, this model simplifies the effective characterization of the long-sequence-related features of SOC. Based on the results of the Beijing Bus Dynamic Stress Test, the proposed SAM-GRU neural network model yields more accurate SOC estimates than the traditional GRU neural network under different discharge rates, environmental temperatures, and combinations of both. Specifically, the improvements in accuracy are no less than 26%, 25%, and 11%, respectively.

Key words: lithium-ion battery, state of charge, self-attentive mechanism, gated recurrent unit neural network

CLC Number: