接头(建筑物)
计算机科学
汽车工程
估计
工程类
噪声测量
可靠性工程
负荷管理
估计理论
控制工程
芯片上的系统
噪音(视频)
电子邮件
数据建模
结构工程
控制理论(社会学)
作者
Lubin Ma,Yuwen Huang,Xuefeng Liu,Bin Duan,Yongzhe Kang,Chenghui Zhang
标识
DOI:10.1109/tii.2026.3736987
摘要
Lithium-ion batteries (LIBs) serve as the core power foundation driving industrial electrification and the new energy transition. The state of charge (SOC) and remaining discharge energy (RDE) directly quantify the relative remaining capacity and absolute deliverable energy of battery systems. Their accurate joint estimation is essential for energy scheduling and safety control, especially for large capacity LIBs. However, wide temperature ranges, nonlinear aging, and complex operating conditions significantly degrade the generalization capability of conventional estimation methods. To this end, this article proposes a semantic-enhanced Mamba (SE-Mamba) framework for accurate joint estimation of SOC and RDE. First, a full-SOC parameter atlas is constructed through parameter identification of a second-order equivalent circuit model, and a virtual tail technique is introduced to generate physically consistent RDE labels. Then, the feature parsing capability of a large language model is leveraged to encode battery signals into a static semantic embedding. This process bridges low-level physical signals and high-level engineering semantics. Based on this, a tandem Mamba network is developed by integrating a pre-estimation network (PreNet) for SOC observation, an online prior generator, and a Mamba backbone. This network structure precisely captures long-sequence nonlinear features with linear computational complexity, establishing solid physical constraints. Experiments across 36 investigated operating condition combinations show that SE-Mamba achieves SOC and RDE mean absolute errors (MAEs) below 0.4% and 1.8 Wh, respectively, outperforming the benchmark methods under the same evaluation protocol.
科研通智能强力驱动
Strongly Powered by AbleSci AI