Physics-Informed Cross-Domain Deep Learning for Laboratory-to-Field Battery Remaining Useful Life Estimation Under Operational Shifts and Target-Label Scarcity
Kumbirayi Nyachionjeka、Emad Abd‐Elrady、Ehab H.E. Bayoumi
论文卡片 2
基本信息
- 标题：Physics-Informed Cross-Domain Deep Learning for Laboratory-to-Field Battery Remaining Useful Life Estimation Under Operational Shifts and Target-Label Scarcity
- DOI/URL：https://doi.org/10.3390/batteries12080274
- 期刊/会议/预印本平台：Batteries
- 发表日期：2026-07-27
- 作者和机构：Kumbirayi Nyachionjeka（Department of Computer and Telecommunications Engineering, University of Botswana）；Emad Abd-Elrady（Crossref 记录含 Faculty of Engineering Technology 字段但截断，完整机构未能确认）；Ehab H.E. Bayoumi（机构未能从可访问截断内容完整确认）。
- 通讯作者：未能确认
- 脚本评分：71
- 期刊指标核验：未核验
纳入理由和命中证据
- 直接任务是锂离子电池 Remaining Useful Life (RUL) estimation，且强调实验室到场景部署迁移。
- Crossref 摘要证据：提出 physics-informed laboratory-to-field (L2F) deep learning framework；六通道循环表示包含 voltage/current/temperature/cumulative charge throughput/cumulative energy throughput/voltage derivative；目标是 battery RUL prediction under limited or unavailable target labels。
创新点、主要内容、方法流程
- 创新点：将物理感知通道、Transformer-TCN 融合主干和无/少目标标签适配策略组合，用于 lab-to-field operational shift。
- 方法流程：
- 构建六通道 laboratory cycle representation；
- 用 gated Transformer-TCN Fusion 建模长程与局部退化模式；
- 通过 paired-view consistency 与 covariance alignment 做 staged adaptation；
- 用 Integrated Gradients 分析物理通道贡献。
输入特征与输出目标
- 输入特征：电压、电流、温度、累计充电吞吐、累计能量吞吐、电压导数。
- 输出目标：RUL 预测轨迹；在 Tsinghua 部署实验中更偏无标签轨迹可信度/平滑性评估。
数据来源/数据结构/电池类型/循环协议
- 电池类型：锂离子电池。
- 数据来源：XJTU laboratory source；Tsinghua University EV field dataset / deployment experiment。
- 数据结构：实验室窗口化循环数据 + 车辆/场景侧无标签目标域轨迹。
- 循环协议、电池数量、化学体系：未能从可访问内容确认。
结果指标、Baseline/对比方法
- 指标：XJTU held-out laboratory RMSE = 46.67。
- Baseline：TCN RMSE = 48.27；Transformer RMSE = 48.85。
- Tsinghua：预处理和窗口构建后没有 measured target RUL labels，因此未计算 field-side MAE/RMSE/R2；报告 operational continuity、finite vehicle-specific predicted trajectories 和局部轨迹波动下降。
- 适配效果：S2a + S2b 将 Fusion RUL volatility 从 8.376 降到 0.632。
- 解释性：physics-aware channels attribution mass 30.98%，早期 19.50%，晚期 31.86%。
泄漏风险/局限性/复现性
- 泄漏风险：中等。优点是作者明确区分有标签 XJTU 与无标签 Tsinghua，不把无标签部署误报为 field-side MAE/RMSE；但需全文确认窗口化、归一化、超参选择是否跨电池隔离。
- 局限性：目标域无 RUL 真值，部署侧只能做轨迹规则性/可信度评估，不能证明实车 RUL 误差。
- 复现性：信息不足；需确认代码、数据处理脚本、XJTU/Tsinghua 数据链接和随机种子。
对哥哥研究/落地的启发和后续动作
- 强烈跟进。该文把“实验室训练模型到真实车辆/储能数据会退化”的问题讲得很具体，尤其适合做域适配 benchmark。
- 后续动作：下载 MDPI PDF；复核通信作者、数据划分、Tsinghua 预处理；把六通道输入和 volatility 指标加入哥哥自己的跨域评估表。
查看关键词分类依据
- RUL prediction · 标题包含「remaining useful life」
eep Learning for Laboratory-to-Field Battery Remaining Useful Life Estimation Under Operational Shifts and Target-Label Scarcity - PINN / physics-informed · 标题包含「physics-informed」
Physics-Informed Cross-Domain Deep Learning for Laboratory-to-Field Battery Remaining Useful Lif - Transformer / attention · 摘要包含「transformer」
throughput, and voltage derivative; a gated Transformer–Temporal Convolutional Network (TCN) Fusion backbone for modeling long-range an - CNN · 摘要包含「convolutional」
age derivative; a gated Transformer–Temporal Convolutional Network (TCN) Fusion backbone for modeling long-range and local degradation pat - Generic neural network / deep learning · 标题包含「deep learning」
Physics-Informed Cross-Domain Deep Learning for Laboratory-to-Field Battery Remaining Useful Life Estimation Under Operatio - Hybrid model · 摘要包含「fusion」
sformer–Temporal Convolutional Network (TCN) Fusion backbone for modeling long-range and local degradation patterns; and a staged a - XJTU · 摘要包含「xjtu」
Fusion backbone achieved the lowest held-out XJTU laboratory root mean square error (RMSE) of 46.67, compared with 48.27 for TCN - Self-built / experimental dataset · 标题包含「laboratory」
sics-Informed Cross-Domain Deep Learning for Laboratory-to-Field Battery Remaining Useful Life Estimation Under Operational Shifts and - Li-ion / LIB · 摘要包含「lithium-ion」
important for the safe and efficient use of lithium-ion (Li-ion) batteries in electric vehicles (EVs) and energy-storage systems. Most