Data-driven prognostics for lithium-ion battery based on Gaussian Process Regression

Abstract

Lithium-ion battery is a promising power source for electric vehicles owing to its high specific energy and power. Through monitoring battery health in effective way such as determining the operating conditions, planning replacement interval could increase the reliability and stability of the whole system. However, due to the reliance on integration, errors in terminal measurement caused by noise, resolution, the uncertainty when we make prognostics for battery health are cumulative, the prediction result is combined with unsatisfied errors. As a result, the prognostic algorithms supporting uncertainty representation and management are emphasized. So in this paper, we present the Gaussian process model to realize the prognostics for battery health. Because of the advantages of flexible, probabilistic, nonparametric model with uncertainty predictions, the Gaussian process model can provide variance around its mean predictions to describe associated uncertainty in the evaluation and prediction. To evaluate the proposed prediction approach, we have executed experiments with lithium-ion battery. Experimental results prove its effectiveness and confirm the algorithm can be effectively applied to the battery monitoring and prognostics. Furthermore, the comparison of prediction with different amounts of training data has been achieved, and the dynamic model is introduced to improve the prediction for the battery health.