计算机科学
芯(光纤)
学习迁移
传输(计算)
控制理论(社会学)
工作(物理)
稳健性(进化)
人工智能
电池(电)
数据传输
理论(学习稳定性)
工程类
钥匙(锁)
噪音(视频)
传递函数
数据收集
控制工程
作者
Xingchen Zhang,Mince Li
标识
DOI:10.1109/tte.2026.3725161
摘要
Core temperature (CT) is a key state variable for ensuring the safe operation of lithium-ion batteries (LIBs) in electric vehicles. In practice, varying ambient temperatures may cause fixed CT estimation method to fail. Existing data-driven CT estimation methods typically need to be retrained or fully recalibrated for each new operating scenario, which is labor-intensive and costly. To overcome this limitation, this paper presents a robust transfer learning framework for CT estimation, which enables efficient adaptation with limited data. Specifically, a CNN–GRU–FC model is first trained at a nominal ambient temperature and used as the pre-trained model. Then, by exploiting the inherent invariance of the battery dynamics, the shared feature-extraction layers are only slightly adjusted, while the output layer is more substantially recalibrated, thereby realizing a differential transfer learning strategy. The proposed framework is validated on LIB experiments under ambient temperatures ranging from -5 °C to 35 °C, demonstrating superior accuracy and robustness compared with conventional retraining and naive fine-tuning approaches, especially when only limited target-domain data are available.
科研通智能强力驱动
Strongly Powered by AbleSci AI