Remaining useful life prediction of lithium-ion batteries using a hybrid model. (1st June 2022)

• Record Type: Journal Article
• Authors: Yao, Fang; He, Wenxuan; Wu, Youxi; Ding, Fei; Meng, Defang
• Is Part Of: Energy. Volume 248(2022)
• Publication Date: 2022-06-01
• Subjects: Lithium-ion battery -- Remaining useful life -- Relevance vector machine -- Extreme learning machine -- Uncertainty expression -- Sensitivity analysis; Power resources -- Periodicals; Power (Mechanics) -- Periodicals; Energy consumption -- Periodicals; 333.7905
• DOI: 10.1016/j.energy.2022.123622
• Languages: English
• ISSNs: 0360-5442

Abstract

Abstract: Accurately predicting the remaining useful life (RUL) of lithium-ion batteries is critical to the stable operation and timely maintenance of a battery system. However, the capacity of an operating battery is difficult to measure, and some prediction models cannot provide an uncertainty expression. To tackle this issue, this paper proposes a hybrid prediction model PSO-ELM-RVM, which integrates particle swarm optimization (PSO), an extreme learning machine (ELM), and relevance vector machine (RVM). Firstly, an indirect health indicator during the constant current charge process is extracted and preprocessed. Secondly, the relationship between the health indicator and capacity is established by RVM, and the health indicator prediction model is constructed based on ELM. PSO is used to optimize the parameters of both the RVM and ELM models. Finally, the health indicator prediction results are added in the RVM model to obtain the predicted capacity with a confidence interval. Compared with the battery failure threshold, the prediction results of RUL can be obtained. The experimental results validate that the proposed model can effectively predict the RUL of lithium-ion batteries. Graphical abstract: Image 1 Highlights: We extract an aging feature related to battery capacity in the charge process. We propose PSO-ELM-RVM model to predict the RUL with an uncertainty expression. The PSO-ELM-RVM employs wavelet denoising algorithm and Box-Cox transformation. Experimental results validate the effectiveness of the proposed model.