A two-layer full data-driven model for state of health estimation of lithium-ion batteries based on MKRVM-ELM hybrid algorithm with ant-lion optimization

Abstract

State of health (SOH) is one of the most important indicators for the lithium-ion batteries' security, reliability and failure, therefore SOH estimation attracts close attention spontaneously. In this paper, a two-layer full data-driven SOH estimation model based on hybrid algorithm composed of multi-kernel relevance vector machine and extreme learning machine optimized with ant-lion optimization (ALO-MKRVM-ELM) is presented. In the model, a pre-estimation layer and an error compensation layer are assembled organically, which use MKRVM algorithm and ELM algorithm respectively. Meanwhile, to solve the problem of tedious debugging for parameters in MKRVM and ELM, ALO algorithm is introduced properly. In addition, considering both of estimation accuracy and calculation complexity, the feature factors for SOH estimation, which can be extracted from the battery's practical operation process, are elaborately selected through correlation analysis also. Finally, the performance comparison against various estimation models was carried out by using two groups of aging experiment datasets from Center for Advanced Life Cycle Engineering (CACLE) and Intelligent Power Laboratory (iPower-Lab) at our university, where CS2-type and ternary lithium-ion batteries were tested respectively, and three statistical evaluation indexes, i.e., the MAE, RMSE, and R2, are applied to assess the estimation results numerically. The experimental results indicate that both accuracy and robustness of the proposed model have been improved significantly.