Future Ageing Trajectory Prediction for Lithium-Ion Battery Considering the Knee Point Effect

Abstract

Lithium-ion batteries have been widely applied in energy conversion sectors, where effective future ageing prediction is crucial to guarantee their safety and performance. Due to the highly nonlinear ageing behaviours, developing a reliable method that could not only consider the knee point effect but also predict the future ageing trajectory with uncertainty quantification poses a formidable task. This paper derives a machine learning solution, based on the migrated Gaussian process regression (GPR), for predicting future battery two-stage ageing trajectory. Specifically, a base model is first offline identified from the easier collected accelerated-speed ageing data, through which the long life ageing information can be effectively learned. With this base model, a migrated mean function is then designed and coupled within the GPR framework for battery ageing predictions. Experimental data from three different batteries are applied for model validation and performance evaluation. Results indicate that the proposed solution leads to effective improvements in prediction accuracy and uncertainty quantification for both cases of training before and after the knee point. This is the first time to couple migration concept within GPR, paving the way to reduce experimental cost and predict battery future two-stage ageing trajectory with only a few (first 30%) data available.