A novel method of battery state-of-health estimation and remaining useful life prediction based on grid search-support vector regression

Abstract

The online estimation of battery SOH (state of health) and prediction of RUL (remaining useful life) are a prerequisite of ensuring its safe and reliable operation, optimal balance and battery SOC estimation. Considering the actual situation, a relatively stable charging environment is more suitable for accurate estimation and prediction of SOH and RUL compared to uncertain discharge conditions. A health indicator based on charging characteristics is selected to predict battery degradation trends through Pearson and Spearman correlation analysis. The grid search optimisation (GS) algorithm and support vector regression (SVR) are integrated to extract the intrinsic relationship between the health indicator and capacity, combining GS algorithm to optimise SVR kernel parameters. With the Lithium-ion data set provided by NASA, it’s proved that the GS-SVR fusion model is more effective in estimating battery life decay than the SVR model, with a reduction of 52.64% in mean absolute error and 68.51% mean square error. Besides, the R² of the GS-SVR model is 97.3%, and the robustness of the model is stronger. The RUL prediction results demonstrate the effectiveness and advantage.

Highlights

• The methods of obtaining indirect health indicators were discussed in detail.

• Data-driven algorithm based on small samples to realise online model estimation.

• The estimation results based on GS-SVR model and SVR model were compared.

• Application to data taken from the Lithium-ion battery cycle aging test of the NASA’s Prognostics Center of Excellence.

KEYWORDS:

• State-of-health
• remaining useful life
• battery health life
• GS-SVR
• data-based
• fusion model

Acknowledgement

We would like to thank the PCoE at NASA for providing the experimental data sets of Li-ion battery.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by Ministry of Science and Technology of the People’s Republic of China [grant number 2017YFB0103600]; Natural Science Foundation of Jiangsu Province [grant number BK20181295].