In-situ data-driven high-precision SOC estimation for silicon-based lithium-ion batteries

Digitalization and intelligentization are pivotal directions for improving the performance and safety of lithium-ion batteries (LIBs) in the future. Focusing on silicon-based lithium-ion batteries, this paper proposes a novel framework for high-accuracy state of charge (SOC) estimation that integrates in-situ sensing with deep learning. Fiber Bragg grating (FBG) sensors are embedded for in-situ thermomechanical data capture, and together with the current and voltage data, they form a multi-dimensional dataset. To address the challenge of SOC estimation accuracy under complex operating conditions, the established dataset includes six dynamic working conditions and two discharge rates, ensuring data diversity. Furthermore, the dataset is subjected to feature engineering (FE) and noise augmentation to enhance the anti-noise capability of SOC estimation. Meanwhile, a hybrid model incorporating a Convolutional Neural Network (CNN), a Gated Recurrent Unit (GRU), and an Attention (Attn) mechanism is developed to simultaneously improve the accuracy and stability of SOC estimation. Experimental results demonstrate that the proposed method reduces the root mean square error (RMSE) of online SOC estimation to 0.635%. The in-situ sensing data contributes significantly to the SOC estimation accuracy, and the combination of diverse operating conditions in the dataset, FE, noise augmentation, and the hybrid model structure collectively guarantees the anti-noise capability and stability of SOC estimation. This research is expected to provide support for the development of artificial intelligence-driven high-precision monitoring technologies for battery states. The dataset is publicly available on Mendeley Data at https://doi.org/10.17632/ft6rtwt8vm.1.