电池(电)
降级(电信)
一致性(知识库)
计算机科学
可靠性工程
子空间拓扑
锂离子电池
编码(内存)
工程类
人工智能
功率(物理)
量子力学
电信
物理
作者
Fujin Wang,Zhibin Zhao,Jiaxin Ren,Zhi Zhai,Shibin Wang,Xuefeng Chen
标识
DOI:10.1016/j.jpowsour.2022.230975
摘要
Predicting Remaining Useful Life (RUL) of lithium-ion batteries plays an important role in battery health management. The existing methods have some limitations in battery RUL prediction, including the problem of error accumulation and the inapplicability of RUL models under different working conditions. For one battery, the existing prediction methods have the problem of error accumulation; for batteries under different working conditions, the corresponding lifetime distributions are discrete, so models cannot be generalized. To address above problems, A transferable RUL prediction method for lithium-ion batteries is proposed. Firstly, an Encoding Net that can extract global degradation tendency information is constructed to map the battery degradation data into the representation subspace. Cycle-consistency learning is used to train the Encoding Net so that points with the same degradation level between different batteries are aligned with each other. Based on aligned features, a new RUL prediction method, consisting of optimal decision of alignment ratio and integrated prediction using Bayesian model averaging (BMA), is established. The experimental results show that the proposed method can achieve high prediction accuracy and also provide a new perspective for RUL prediction.
科研通智能强力驱动
Strongly Powered by AbleSci AI