Aging-Aware Classification and Optimal Usage of Electric Vehicle Batteries

Doktorsavhandling, 2025

To facilitate successful market adoption of second-life battery energy storage systems (BESSs) based on new and used electric vehicle (EV) batteries, we propose aging-aware classification in first-life applications and optimal usage in second-life applications of EV batteries.

Specifically, a transferable physics-informed framework is proposed for battery degradation mode estimation and phase detection, and a quantile regression forests (QRF) model is proposed for battery lifetime early prediction, which enables aging-aware classification of EV batteries. For optimal usage of EV batteries in second-life BESS applications, an economic stage cost function is proposed to account for both the grid and the battery degradation cost, and an automatic kernel search method is extended to construct the best composite kernel for Gaussian process (GP) regression models in two battery applications.The fine-tuning strategy is proven to be effective in improving online battery degradation mode estimation and phase detection performance in the target first-life application. The QRF model can provide cycle life point prediction with high accuracy, and uncertainty quantification as the width of prediction intervals. The implicit policy incorporating historical operational data and "fixed" forecasted electricity price achieves the best economic performance, and GP regression models with the best kernel can provide better prediction performance in the two battery applications.

In summary, machine learning methods are proposed in this thesis to enable aging-aware classification in first-life applications and optimal usage in second-life applications of EV batteries, which hopefully facilitates the market adoption of second-life BESSs based on new and used EV batteries.