Data-driven prognostic techniques for estimation of the remaining useful life of Lithium-ion batteries

This paper aims to study the use of various data-driven techniques for estimating the remaining useful life (RUL) of the Li-ion batteries. These data-driven techniques include neural networks, group method of data handling, neuro-fuzzy networks, and random forests as an ensemble-based system. These prognostic techniques make use of the past and current data to predict the upcoming values of the capacity to estimate the remaining useful life of the battery. This work presents a comparative study of these data-driven prognostic techniques on constant load experimental data collected from Li-ion batteries. Experimental results show that these data-driven prognostic techniques can effectively estimate the remaining useful life of the Li-ion batteries. However, the random forests and neuro-fuzzy techniques outperform other competitors in terms of the RUL prediction error and root mean square error (RMSE), respectively.