Taylor polynomial-constrained physics-informed neural network for battery degradation modeling and SOH estimation
Wenlan Xu
论文卡片 5
基本信息
- 标题:Taylor polynomial-constrained physics-informed neural network for battery degradation modeling and SOH estimation
- DOI/URL:https://doi.org/10.1088/1361-6501/ae83d4
- 期刊/会议/预印本平台:Measurement Science and Technology
- 发表日期:2026-06-29(Accepted Manuscript online)
- 作者和机构:Wenlan Xu;Chongqing University, No. 55, South University Town Road, Chongqing High-Tech Zone, Chongqing, 400044, China
- 通讯作者:Wenlan Xu(IOP 页面列出 Email: 20230025@stu.cqu.edu.cn;单作者文章,通讯角色未以“corresponding author”明示,但可联系邮箱为其本人)
- 脚本评分:61
- 期刊指标核验:未核验。
纳入理由和命中证据
摘要明确写明 SOH estimation、battery degradation modeling、battery degradation dynamics,并提出 Taylor-PINN 用于 SOH 估计。主题符合电池健康预测。
创新点
- 将 Taylor polynomial 嵌入 PINN,用 Taylor residual 约束相邻循环间的平滑退化动力学。
- 引入 monotonicity constraint 抑制不物理的容量回弹。
- 用 parallel LSTM 分别从 CC voltage data 与 CV current data 提取健康特征。
- 在 XJTU 数据集与 TJU 数据集上验证,强调跨化学体系/工况泛化。
主要内容
IOP accepted manuscript 页面和摘要显示,该研究试图把物理一致性约束与深度学习结合,解决纯数据驱动缺乏物理解释、物理模型依赖先验参数的问题。
模型架构/方法流程
流程为:charging process health feature selection → P-LSTM 从 CC 电压和 CV 电流提取 HFs → Taylor-PINN degradation model → Taylor residual constraint + monotonicity constraint + data fitting joint loss → SOH 输出。完整网络层数、超参和训练细节未能从可访问页面确认。
输入特征与输出目标
- 输入特征:constant-current voltage data、constant-voltage current data 中的健康特征。
- 输出目标:SOH / degradation trajectory。
数据来源/数据结构/电池类型/循环协议
- 数据集:XJTU dataset;TJU battery dataset。
- 电池类型:Lithium-ion batteries;摘要称跨 cathode chemistries and operating conditions。
- 具体电芯数量、温度/倍率/协议:未能从摘要页面确认。
结果指标
摘要称 Taylor-PINN 在不同 batteries 与 charge-discharge protocols 下取得 lowest MAE and RMSE,并有 ablation studies;具体数值未能确认。
Baseline/对比方法
摘要只称 outperform other SOTA models;具体 baseline 名称未能从可访问页面确认。
泄漏风险/局限性/复现性
- 泄漏风险:中等。SOH 估计论文常见风险是循环片段随机切分和归一化泄漏;需全文确认。
- 局限性:IOP 页面提示当前访问未订阅 Version of Record,只能看到 accepted manuscript 摘要级信息。
- 复现性:信息不足;未确认代码/数据链接。
对哥哥研究/落地的启发和后续动作
- 可跟进 Taylor residual + monotonicity constraint,作为容量轨迹模型的正则项。
- 后续动作:如果需要 PINN 方向,优先核验其数学形式和 ablation 贡献,避免只停留在“物理约束”概念层。
查看关键词分类依据
- SOH estimation · 标题包含「soh」
network for battery degradation modeling and SOH estimation - PINN / physics-informed · 标题包含「physics-informed」
Taylor polynomial-constrained physics-informed neural network for battery degradation modeling and SOH estimation - LSTM · 摘要包含「lstm」
bility. A parallel long short-term memory (P-LSTM) network is then designed to extract HFs from the constant-current (CC) voltage - Generic neural network / deep learning · 标题包含「neural network」
ylor polynomial-constrained physics-informed neural network for battery degradation modeling and SOH estimation - Bayesian · 摘要包含「uncertainty」
charging process is introduced, eliminating uncertainty introduced by discharge variability. A parallel long short-term memory (P-LSTM) - Hybrid model · 摘要包含「fusion」
t data, capturing polarization growth and diffusion-related aging information, respectively. Finally, a degradation model is constr - XJTU · 摘要包含「xjtu」
cal consistency. Experimental results on the XJTU dataset demonstrates that the proposed Taylor-PINN outperforms other SOTA model - Self-built / experimental dataset · 摘要包含「experimental」
stimation accuracy and physical consistency. Experimental results on the XJTU dataset demonstrates that the proposed Taylor-PINN outperfo - Li-ion / LIB · 摘要包含「lithium-ion」
essential for the safety and reliability of lithium-ion batteries. The battery degradation dynamics are influenced by electrode materia