A novel multi-scenario battery health assessment method combining semi-supervised learning and data augmentation techniques

Data-driven methods are widely claimed to be the most promising candidates for online battery health assessment estimation. However, sufficient training data cannot be guaranteed. This paper proposes a novel multi-scenario battery health assessment method. First, an efficient feature extraction method that requires no complex calculation is proposed. Besides, the selected features are proven to be temperature independent. Second, a battery data augmentation approach is proposed to enrich unlabeled battery data. Third, different health estimation strategies are applied for different scenarios. For supervised scenarios, k -nearest neighbor is directly used for health assessment. For semi-supervised scenarios, back propagation neural network and k -nearest neighbor are combined together to overcome the overfitting problem of the former and the poor prediction ability of the latter. In cases where real unlabeled data is not available, the proposed data augmentation method is used to enrich the dataset. The results indicate that the proposed method not only achieves high-precision estimation in fully supervised scenarios, but also significantly improves estimation accuracy in semi supervised scenarios through the combination of two algorithms. The proposed data augmentation method has also been proven to synthesize data that is indistinguishable from real data.