Innovative multiscale fusion - antinoise extended long short-term memory neural network modeling for high precision state of health estimation of lithium-ion batteries.

Tao, Junjie; Wang, Shunli; Cao, Wen; Cui, Yixiu; Fernandez, Carlos; Guerrero, Josep M.

Abstract

An accurate assessment of lithium-ion (Li-ion) batteries' state of health (SOH) is essential for the safe operation of new energy systems and extended battery life. Health factors were extracted by studying the aging test data of Li-ion batteries to estimate the health state. A multi-scale data fusion and anti-noise extended long short-term memory (LSTM) neural network is proposed. The current, voltage, and other micro-scale data of Li-ion batteries were extracted by fast Fourier transform (FFT), and the main frequency characteristics were extracted by principal component analysis (PCA). The hidden layer structure of the LSTM neural network is extended to separate independent positive and negative correlation gating weight parameters to reduce the risk of overfitting. At the same time, a novel network weight updating algorithm combining an extended Kalman filter (EKF) and gradient descent (GD) is proposed, and the inherent noise suppression property of the EKF is utilized to improve the algorithm's robustness. The experimental results show that the accuracy of the MSDF-ANELSTM algorithm is improved by 66.66%, stability by 83.84%, and generalization performance by 72.54% compared with the traditional neural network. This is conducive to promoting the industrial application of data-driven Li-ion battery management systems.

Citation

TAO, J., WANG, S., CAO, W., CUI, Y., FERNANDEZ, C. and GUERRERO, J.M. 2024. Innovative multiscale fusion - antinoise extended long short-term memory neural network modeling for high precision state of health estimation of lithium-ion batteries. Energy [online], 312, article number 133541. Available from: https://doi.org/10.1016/j.energy.2024.133541

• Journal Article Type: Article
• Acceptance Date: Oct 19, 2024
• Online Publication Date: Oct 20, 2024
• Publication Date: Dec 15, 2024
• Deposit Date: Oct 21, 2024
• Publicly Available Date: Oct 21, 2025
• Journal: Energy
• Print ISSN: 0360-5442
• Electronic ISSN: 1873-6785
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Volume: 312
• Article Number: 133541
• DOI: https://doi.org/10.1016/j.energy.2024.133541
• Keywords: Lithium-ion battery health state estimation; Fast Fourier transform; Principal component analysis; Multi-scale data fusion; Anti-noise extended LSTM neural network
• Public URL: https://rgu-repository.worktribe.com/output/2541351