Early Prediction of Lithium-Ion Battery Cycle Life by Machine Learning Methods

For a lithium-ion battery, its cycle life is defined as the number of full charge cycles that a battery can undergo until its full charge capacity falls below 80% of the design capacity. In a recent study by Severson et al. [1], a large set of lithium-ion battery cycle life experiments were conducted and analyzed, and early cycle data were used to predict battery lives without any prior assumptions on degradation mechanism. In this paper, we reexamine the data and suggest additional features to the model, which also use early cycle data (up to the first 100 cycles), for a better battery cycle life prediction. We suggest an ensemble machine learning method that combines several classifiers such as the k-nearest neighbor classifier, neural networks, support vector machines and decision tree-based classifiers for classifying batteries to low or high lifetime category. It is found that our ensemble approach provides more robust predictions. For predicting cycle life, a support vector regression model is suggested, and we compare it with an elastic net-based regression model. It is found that SVR outperforms elastic net in terms of percentage prediction error.