Real-Time Capacity Estimation of Lithium-ion Batteries Using a Novel Ensemble of Multi-Kernel Relevance Vector Machines

Abstract

Lithium-ion batteries have been growing in popularity for portable electronics, electric vehicles, aerospace and military devices due to many excellent characteristics. The prognostics and health management of lithium-ion batteries are significant. In this paper, a novel mixture model of multi-kernel relevance vector machines with dynamic weights (DW-MMKRVM) is proposed to estimate the real-time capacity of lithium-ion batteries based on indirect health indicators. Weights of each sub-model in DW-MMKRVM keep updating during sequential, online data collection and model training. Experiments illustrate the proposed approach can produce more robust and accurate capacity estimation, which is critical for prognostics and health management of lithium-ion batteries. Comparison results also show that the proposed DW-MMKRVM with more sub-models can increase the estimation accuracy.