学习迁移
计算机科学
人工智能
卷积神经网络
电池(电)
深度学习
均方误差
特征(语言学)
机器学习
人工神经网络
健康状况
模式识别(心理学)
功率(物理)
统计
数学
哲学
物理
量子力学
语言学
作者
Penghua Li,Yi Cheng,Kangheng Shan,Yifan Fang
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2022-08-15
卷期号:: 889-894
被引量:6
标识
DOI:10.1109/ccdc55256.2022.10033695
摘要
For overcoming the data scarcity faced in state-of-health (SOH) estimation of lithium-ion batteries, this paper proposes a novel, hybrid, deep-transfer learning strategy called CNN-BiLSTM-MMD. Combining convolutional neural network (CNN) with bidirectional long short-term memory (BiLSTM) neural network, a CNN-BiLSTM is created to be the base model of CNN-BiLSTM-MMD. The knowledge learned from relevant battery data is transferred into the CNN-BiLSTM through feature transfer learning based on the maximum mean difference (MMD). Our method can make up for the lack of information caused by the insufficient training dataset, and realize high-precision lithium-ion battery health monitoring based on limited historical measurements. Compared to some existing data-driven methods and parameter-based transfer learning, the experiments carried on the NASA dataset demonstrate our approach hits lower average root mean square error(ARMSE), 0.01469, in the case of the limited training dataset.
科研通智能强力驱动
Strongly Powered by AbleSci AI