Time-efficient identification of lithium-ion battery temperature-dependent OCV-SOC curve using multi-output Gaussian process

For lithium-ion batteries, the functional dependence of open circuit voltage (OCV) on state of charge (SOC) varies with temperature and aging, which plays a significant role in accurate SOC estimation and state of health monitoring. To identify the OCV-SOC curve at a given condition, OCVs usually need to be either measured by a time-consuming OCV test, or estimated with inevitable errors that eventually propagate into the identified OCV-SOC curve. In this paper, we investigate time-efficient identification of temperature-dependent OCV-SOC curve from current-voltage data, without measuring or estimating OCVs. In particular, we identify the complete OCV-SOC curve from data over a partial SOC range at a given temperature, by fusing available OCV-SOC curve data at other temperatures. In the proposed approach, a multi-output Gaussian process (MOGP) model is first built to capture correlations among OCV-SOC curves at different temperatures, and then used to construct the OCV-SOC curve at the given temperature. Using experimental datasets, our proposed approach reduces the root mean square error (RMSE) of OCV predictions by at least 29.4% compared to three existing methods. Besides, with the updated OCV-SOC curve, the RMSE of SOC estimates is reduced by at least 14.0%, compared to using a non-updated OCV-SOC curve.

Automated summary

In this paper, a time-efficient method is proposed to identify the temperature-dependent OCV-SOC curve from current-voltage data by fusing existing OCV-SOC curve data at different temperatures. Experimental results show that our approach significantly reduces the RMSE of OCV predictions by at least 29.4% compared to existing methods, and also improves the accuracy of SOC estimates by at least 14.0% using the updated OCV-SOC curve.