Hierarchical Clustered Federated Learning for Battery State-of-Health Estimation Privacy-Preserving Personalized Prediction with Feature Decoupling and Edge Deployment
Abstract
This thesis addresses the issues of data heterogeneity, privacy protection, and distributed modeling in lithium-ion battery state-of-health (SOH) estimation, proposing a hierarchical clustering federated learning framework for battery SOH prediction. Traditional centralized learning methods typically rely on uploading raw operational data from different vehicles or battery systems to the cloud. However, in practical applications, battery data is often distributed across different devices or institutions and is affected by differences in temperature, charging/discharging strategies, usage habits, and aging stages. Therefore, how to achieve cross-client collaborative SOH modeling without sharing raw data is the core issue of this research. First, this thesis establishes a unified SOH label definition and multi-source heterogeneous data processing flow based on three public battery datasets: NASA PCoE, MIT Battery Cycle Life, and HUST. Then, combining battery aging mechanisms and statistical feature analysis, the input features are divided into a shared degradation feature stream and a local context feature stream. In terms of model design, this thesis constructs a temporal prediction network based on LSTM, Adapter, and Attention, and further builds a personalized federated learning structure based on shared representations for edge-end private feature correction. The shared module participates in federated aggregation, while the private residual branch remains local to the…
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.