Accurate estimation of battery state of health (SOH) during fast charging is crucial for the management of electric vehicle batteries. However, challenges remain due to the lack of training data for individual target batteries and the need for personalized models to account for variations in charging and discharging behavior.
In a study published in IEEE Transactions on Transportation Electrification, a research team led by Prof. CHEN Zhongwei and Assoc. Prof. MAO Zhiyu from the Dalian Institute of Chemical Physics (DICP) of the Chinese Academy of Sciences (CAS), in collaboration with Prof. FENG Jiangtao from Xi'an Jiaotong University, developed a novel two-stage federated transfer learning framework for accurate SOH prediction using fast charging segments while preserving user privacy.
The proposed FTL framework for fast charging battery SOH estimation (Image by LIU Yunpeng)
In this framework, multiple distributed batteries first collaborate to train a global model through federated learning, sharing only model parameters. This allows the system to learn general battery behavior without exposing private data. The global model is then fine-tuned with a small amount of local data from a target battery, generating a personalized model that captures its unique charging and discharging characteristics.
The framework is built on a lightweight convolutional neural network enhanced with a channel attention mechanism. The Experimental results on the public fast charging battery dataset show that it outperforms both locally trained models and traditional federated learning methods.
Additionally, this federated transfer framework has been integrated into the second-generation battery digital brain (PBSRD Digit core model), enabling intelligent battery management. It has been applied to launch a vertical intelligent customer service system in the energy storage field for Shuangdeng Group, advancing automation and intelligence in the energy storage industry.
"This federated transfer learning technology provides solid technical support for our intelligent customer service system," said Prof. CHEN.