Prediction of Lithium-ion Battery's Remaining Useful Life Based on Multi-kernel Support Vector Machine with Particle Swarm Optimization

The Remaining Useful Life (RUL) of Lithium-ion (Li-ion) batteries estimation is very important for the intelligent Battery Management System (BMS). With the arrival of the era of big data, data mining technology is becoming more and more mature, and the Li-ion batteries RUL estimation based on data-driven prognostics is more accurate. However, the Support Vector Machine (SVM) applying to predict the Li-ion batteries RUL uses the traditional single radial basis kernel function, this single radial basis kernel function classifier is weak in generalization ability, and it is easy to appear the problem of data migration, which leads to the inaccurate prediction of Li-ion batteries RUL. In this work, a novel Multi-kernel SVM (MSVM) based on polynomial kernel and radial basis kernel function is proposed. Moreover, the Particle Swarm Optimization (PSO) algorithm is used to search the kernel parameters, penalty factor and weight coefficient of the MSVM model. Finally, this paper makes use of the NASA battery data set to form the observed data sequence for regression prediction, the calculation results show that the improved algorithm not only has better prediction accuracy and stronger generalization ability, but also decreases the training time and computational complexity.