FedCBE: A federated-learning-based collaborative battery estimation system with non-IID data

State of Charge (SOC) estimation of lithium-ion batteries using deep learning methods has attracted considerable attention. However, existing approaches predominantly rely on centralized learning paradigms, necessitating datasets generated under diverse conditions. This process can be time-consuming and even impractical for real-world applications due to data privacy. In this study, we propose a Federated-Learning-Based Collaborative Battery Estimation system (FedCBE) for SOC estimation of lithium-ion batteries. The proposed FedCBE leverages federated learning (FL), which requires only local model weights rather than raw data for global model aggregation, thereby enhancing the model's training efficiency and mitigating data privacy concerns. Additionally, we propose label normalization, proximal term, and shared-data strategies to avoid weight divergences encountered during FL training, thus improving the stability and generalization of the proposed FedCBE. Both open-source and our laboratory battery datasets are employed to demonstrate the efficacy of the FedCBE. The results reveal that the global model trained on open-source datasets achieves RMSE under 4.5% on our laboratory battery datasets.