Application of Ensemble Learning for Improving Failure Prediction in Lithium-Ion Batteries

Machine Learning (ML) models suffer from bias and over-fitting and have dire results for some applications. In order to soften this issue, approaches like Ensemble Learning (EL) are used to give higher accuracy and robustness for several classification and regression problems. One of them can be the failure detection in lithium-ion batteries (LIBs). This technology is the current trend for mobility and intelligent grids but can suffer from failures if it does not operate under safe and controlled conditions. Therefore, this paper applied 5 EL advanced techniques to demonstrate the efficiency of meta-learning to improve failure detection in LIBs by predicting the temperature of the cells. The results demonstrated that they significantly reduced the error compared to single learners, especially in the stacking, bagging, and blending approach. Despite this application, this work demonstrated that EL approach could improve the performance for several problems in the industry.