The particle swarm optimization multi-kernel relevance vector machine for remaining useful life prediction of lithium-ion batteries

• Published in: IEEE Advanced Information Technology, Electronic and Automation Control Conference (IAEAC) (Online) pp. 938 - 943
• Main Authors: Tian, Qiancheng, Liu, Xiping, Ding, Shuai, Chen, Haitao, Huang, Jun, Yang, Ziwei
• Format: Conference Proceeding
• Language: English
• Published: IEEE 03.10.2022
• Subjects: Degradation; Lithium-ion batteries; Lithium-ion battery; Particle swarm optimization algorithm; Prediction algorithms; Relevance vector machine; Relevance vector machines; Remaining useful life prediction; Space vehicles; Support vector machines; The phase of space reconstruction; Training
• ISSN: 2689-6621
• DOI: 10.1109/IAEAC54830.2022.9929806

The long life of spacecraft has put forward higher requirements for the prediction of the remaining useful life (RUL) of the lithium-ion battery, and the prediction method based on relevance vector machine (RVM) has also received extensive attention. The kernel function is the main factor affecting the prediction of the RVM. Single-kernel RVM has only one kernel function, and the RUL prediction effect is ordinary. Multi-kernel RVM has many types of kernel functions, but the weight coefficient of the kernel function is relatively difficult to determine. Therefore, this paper proposes a multi-kernel RVM based on a particle swarm optimization (PSO) algorithm to predict the RUL of the lithium-ion battery. Firstly the capacity degradation data of the lithium-ion battery is reconstructed in phase space. Then Gaussian, polynomial, Sigmoid, and linear kernel functions are used to establish a multi-kernel RVM model. Finally, the particle swarm optimization algorithm is used to perform parameter self-optimization. Taking the lithium-ion battery capacity degradation data set as an example, according to the prediction evaluation index, the PSO multi-kernel RVM prediction method is better than the single-kernel and multi-kernel RVM prediction method. This method increases the types of kernel functions, solves the problem of difficulty in determining the weight coefficients of the kernel function, and improves the accuracy of predicting the RUL of the lithium-ion battery.