C2L-Net: A Data-Driven Model for State-of-Charge Estimation of Lithium-Ion Batteries During Discharge
Researchers have introduced C2L-Net, a novel data-driven framework designed to enhance the accuracy and efficiency of State-of-Charge (SOC) estimation for lithium-ion batteries within Battery Management Systems (BMS). Addressing the limitations of existing methods that rely on long historical sequences and suffer from high computational costs, C2L-Net utilizes a short 20-second historical window. The model explicitly separates contextual encoding from latest-measurement updating, employing a chunk-based feature extraction mechanism with Theta Attention Pooling and a Fourier-based Seasonality Basis. This architecture includes a causal context encoder integrating Gated Recurrent Units (GRU) with Causal Cosine Attention to prevent information leakage, alongside a decoder inspired by recursive filtering for rapid adaptation to dynamic conditions. Experimental results on public datasets demonstrate that C2L-Net achieves state-of-the-art accuracy while offering up to 60 times faster inference speeds and requiring fewer parameters than current baselines. This advancement significantly improves computational efficiency and robustness across unseen driving profiles, marking a substantial step forward in real-time battery management technology.
Wire timeline
C2L-Net: A Data-Driven Model for State-of-Charge Estimation of Lithium-Ion Batteries During Discharge
Researchers have introduced C2L-Net, a novel data-driven framework designed to enhance the accuracy and efficiency of State-of-Charge (SOC) estimation for lithium-ion batteries within Battery Management Systems (BMS). Addressing the limitations of existing methods that rely on long historical sequences and suffer from high computational costs, C2L-Net utilizes a short 20-second historical window. The model explicitly separates contextual encoding from latest-measurement updating, employing a chunk-based feature extraction mechanism with Theta Attention Pooling and a Fourier-based Seasonality Basis. This architecture includes a causal context encoder integrating Gated Recurrent Units (GRU) with Causal Cosine Attention to prevent information leakage, alongside a decoder inspired by recursive filtering for rapid adaptation to dynamic conditions. Experimental results on public datasets demonstrate that C2L-Net achieves state-of-the-art accuracy while offering up to 60 times faster inference speeds and requiring fewer parameters than current baselines. This advancement significantly improves computational efficiency and robustness across unseen driving profiles, marking a substantial step forward in real-time battery management technology.
cs.AI updates on arXiv.org