Online data-driven battery life prediction and quick classification based on partial charging data within 10 min

Accurate online battery life prediction is critical for the health management of battery powered systems. This study develops a moving window-based method for in-situ battery life prediction and quick classification. Five features are extracted from the partial charging data within 10 min to indicate battery aging evolution. The machine learning techniques are used to connect the features and battery end of life (EOL), with the Gaussian process regression (GPR) and support vector machine (SVM) used to predict and classify battery life, respectively. The performance of the developed methods is validated based on experimental data of 121 battery cells. Results show that GPR predicts accurate battery EOL and knees with the root mean square errors and mean absolute percentage errors being within 100 cycles and 10 %, respectively. SVM classifies battery life quickly and accurately based on only one cycle's data, with the classification accuracy close to 92 %. In summary, the developed methods show comparable in-situ life prediction and classification accuracies to the benchmark that needs offline calibrations. •A partial charging-based moving window method is used to capture battery aging.•A novel physical feature is extracted from partial charging data within 10 min.•The uncertainties of battery life and knee point prediction are quantified.•Life prediction results are comparable to the benchmark using whole discharge data.•Battery life is classified quickly and accurately based on onboard data of 1 cycle.