Battery Life Early Prediction with Feature Optimization

Battery life early prediction is significant to battery development and management. However, how to effectively select reasonable and efficient battery features for early prediction is still a big challenge. This paper proposes a novel method of early battery life prediction with feature optimization. Firstly, some ineffective and redundant battery features are filtered based on the mutual information between battery features and battery life, and correlation coefficient between battery features. Then, considering the complex nonlinearity of the battery capacity decline and the small amount of experimental data, kernel principal component analysis is adapted to reorganize the retained features to make the data more linear and differentiable. Finally, based on the reconstructed data, the early prediction model of battery life is obtained by using Gaussian process regression. The experiments on Severson’s dataset show the effectiveness of the proposed feature optimization method.