A unified GPR model based on transfer learning for SOH prediction of lithium-ion batteries

Abstract

State of health (SOH) acts as a qualitative capability measure in lithium-ion batteries’ management systems. Accurate SOH prediction is a critical issue for lithium-ion batteries. Most existing techniques always extract features from the tested batteries’ historical charging/discharging curves to achieve SOH prediction. However, the charging or discharging curves may be incomplete in the real-world application. Also, it is necessary to provide effective and dependable SOH predictions for both one-step-ahead and multi-step-ahead scenarios simultaneously, catering to diverse requirements. In order to achieve a unified SOH prediction without a prediction lag, a Gaussian process regression (GPR) model based on transfer learning is proposed. In this article, a non-zero mean function along with a compound covariance function are designed to describe the capacity attenuation. The hyper-parameter set of this model can be transferred and pre-determined from some readily available batteries in the same processes. The proposed method is verified on several batteries from NASA dataset. Results illustrate that our approach with both superior prediction performance and stronger robustness outperforms the counterparts.