Early prediction of battery lifetime for lithium-ion batteries based on a hybrid clustered CNN model

Early prediction of lithium-ion battery lifetime is critical for energy storage equipment, because it can provide users with early warnings and alerts to avoid potential disasters. However, making an accurate early prediction is challenging due to the negligible capacity degradation and the scarcity of data in the early stages. To address this issue, a hybrid clustered CNN model is proposed for early prediction of the battery cycle life, which uses both manually extracted and machine learned features as input to strengthen the degradation characterization ability. Additionally, battery classification is innovatively utilized to boost prediction performance. After classifying the batteries into “short cycle life” and “long cycle life”, two improved CNN-based models are accordingly put forward for the two categories, resulting in a significant promotion in prediction accuracy. A well-known public dataset is employed for validation. Comparison results demonstrate that the proposed model exhibits superior prediction accuracy with a relatively small amount of training data. Meanwhile, an additional dataset is used to verify the repeatability and generalizability of the proposed approach. This study reveals a new perspective that the preliminary classification can contribute to the early prediction of battery lifetime.