State of health prediction for lithium-ion battery using a gradient boosting-based data-driven method
| Autor: | Zhiyuan Liu, Linhui Zhao, Pengliang Qin |
|---|---|
| Rok vydání: | 2022 |
| Předmět: |
Battery (electricity)
Boosting (machine learning) Renewable Energy Sustainability and the Environment business.industry State of health Computer science Energy Engineering and Power Technology Machine learning computer.software_genre Data-driven Tree (data structure) symbols.namesake Taylor series symbols Gradient boosting Artificial intelligence Electrical and Electronic Engineering business computer Voltage |
| Zdroj: | Journal of Energy Storage. 47:103644 |
| ISSN: | 2352-152X |
| DOI: | 10.1016/j.est.2021.103644 |
| Popis: | Accurate SOH (State of Health) prediction of lithium-ion battery is of great significance for battery maintenance and safe driving of electric vehicles. To obtain the excellent SOH prediction results, this paper first proposes a novel method for extracting aging features based on the shape of voltage curve. Then, based on the idea of gradient boosting, a novel gradient boosting-based data-driven method is proposed from the selection of the initial learning machine, the second-order Taylor expansion of the negative gradient of the loss function and the setting of the learning rate. The proposed method can continue to learn and reduce errors based on the prediction results of existing data-driven algorithms. Finally, to make the proposed gradient boosting-based data-driven method satisfy the needs of online learning and prediction, a Hoeffding tree-based online incremental learning strategy is designed. The experimental results demonstrate that the selected aging features are reasonable and effective, the proposed gradient boosting-based data-driven method can availably improve the prediction accuracy of the data-driven algorithm, the boosting effect can be increased by up to 78.87%. The designed online incremental learning strategy can reduce the learning time of the proposed algorithm by 10 to 50 times, which is easier to meet the needs of SOH online learning and prediction. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full Text from ScienceDirect