Lightweight Feature-Based Attention Network for Li-Ion Battery SOC Estimation

Accurate state of charge (SOC) estimation is crucial for the efficient operation of electric vehicle batteries, particularly across varying temperature ranges, as it extends battery life and endurance. While deep learning has shown promise in SOC estimation, its computational overhead remains a challenge. Developing a lightweight, efficient, and accurate deep learning SOC estimator is key to deployment in battery management systems (BMS). This paper addresses this challenge by proposing a novel feature-based attention deep learning model for SOC estimation. The model is trained and evaluated on the LG 18650HG2 Li-ion dataset under diverse driving scenarios and battery conditions. Its robustness is assessed by introducing current noise into the input, and its computational efficiency is analyzed on a microprocessor to simulate its workload in a BMS, demonstrating its real-world deployability. The proposed model achieves an RMSE of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1. 2 3 \%}$</tex> on the test dataset with only 1,713 parameters, highlighting its performance and efficiency for BMS deployment. The code and dataset used in this study are available at https://github.com/ahmedsalahacc/efficient-feature-attention-SoC-estimation.