An improved convolutional neural network-bidirectional gated recurrent unit algorithm for robust state of charge and state of energy estimation of new energy vehicles of lithium-ion batteries.

Wu, Fan; Wang, Shunli; Liu, Donglei; Cao, Wen; Fernandez, Carlos; Huang, Qi

Abstract

State of charge (SOC) and state of energy (SOE) are the key factors that reflect the safe and range driving of new energy vehicles. This paper proposes an optimized convolutional neural network-bidirectional gate recurrent unit (CNN-BiGRU) and an improved Kalman bidirectional smoothing algorithm to predict SOC and SOE accurately. Firstly, the attention mechanism (AM) is introduced into the CNN-BiGRU to extract significant features better. A multi-task learning (MTL) mechanism is constructed to learn the correlation between tasks and output the results simultaneously. Then, square root and reverse smoothing methods are added to the classical extended Kalman filtering (SREKS) to filter and de-noise the neural network output. Finally, the test data of the hybrid pulse power characterization (HPPC) and Beijing bus dynamic stress test (BBDST) at 15 °C and 35 °C are used for experimental verification. Under the HPPC condition, the maximum error of SOC and SOE is less than 0.01432 and 0.01417, respectively. Under BBDST condition, the maximum error of SOC and SOE is less than 0.01827 and 0.01883, respectively. Experimental results show that this algorithm has high accuracy and robustness under different complex working conditions.

Citation

WU, F., WANG, S., LIU, D., CAO, W., FERNANDEZ, C. and HUANG, Q. 2024. An improved convolutional neural network-bidirectional gated recurrent unit algorithm for robust state of charge and state of energy estimation of new energy vehicles of lithium-ion batteries. Journal of energy storage [online], 82, article 110574. Available from: https://doi.org/10.1016/j.est.2024.110574

• Journal Article Type: Article
• Acceptance Date: Jan 11, 2024
• Online Publication Date: Jan 18, 2024
• Publication Date: Mar 30, 2024
• Deposit Date: Feb 1, 2024
• Publicly Available Date: Jan 19, 2025
• Journal: Journal of energy storage
• Print ISSN: 2352-152X
• Electronic ISSN: 2352-1538
• Publisher: Elsevier
• Peer Reviewed: Peer Reviewed
• Volume: 82
• Article Number: 110574
• DOI: https://doi.org/10.1016/j.est.2024.110574
• Keywords: Convolutional neural network; Bidirectional gate recurrent unit; Attention mechanism; Multi-task learning mechanism; Square root extended Kalman filter; Lithium-ion battery
• Public URL: https://rgu-repository.worktribe.com/output/2225784