Research on the Prediction Technique for Battery Performance Degradation Trends Based on Hierarchical Federated Learning

Lithium-ion batteries have a wide range of applications as a new energy source, and it is of great significance to realize the prediction of the performance degradation trends of lithium-ion batteries. At present, data-driven prediction technology for performance degradation trends has become an effective method for the prediction of lithium batteries. However, in the actual prediction process, lithium batteries have less data available for training, resulting in poor training accuracy. In this paper, we propose a prediction algorithm for battery performance degradation trends based on hierarchical federated learning to improve the accuracy of lithium battery prediction. First, we propose a federal grouping technique based on improved K-medoids clustering to group multi-formulation batteries. Then, we perform hierarchical federated learning based on the grouping results. Com pared with traditional client-server federation learning, hierarchical federation learning reduces the server communication burden and improves the model convergence speed. In addition, we use a multiscale time convolutional network prediction method to predict a single cell. The method involves predicting trends at both large and small scales, followed by a refinement process to enhance the accuracy of the predictions. The method proposed in this paper has good performance on the dataset.