A Bayesian Predictive Inference with Parametric Regression Model for Battery SOH and RUL Predictions

Effective prognostic and health management of lithium-ion battery-powered systems relies heavily on accurate prediction of battery internal state parameters, including state of health (SOH) and remaining useful life (RUL). However, most existing models for predicting SOH and RUL are either computationally demanding due to their high complexity or suffer from poor adaptability for new battery cells with limited data. To address these challenges, this paper proposes a Bayesian Predictive Inference with a Parametric Regression (BIPR) model for robust long-term SOH and RUL predictions. The proposed model is designed to leverage knowledge transferred from aged battery cells and adapt it to a hypothetical new battery cell, which is still at its early stage of life. The proposed predictive framework starts by training a highly accurate base model using aged battery cells with full life cycle data. It then construct a prior distribution and define the likelihood, thereafter, the posterior is updated for prediction according to the available limited data of the new battery through Bayesian updating procedure. Experimental results show that the proposed method outperforms existing state-of-the-art benchmark methods, this is partly because of its relatively greater generalization ability.