A Novel Lithium-ion Battery State of Health Estimation Model: Integrating Transfer Learning with Retentive Network

Lithium-ion batteries are increasingly critical in portable and electrical technologies, making accurate estimations of their State of Health essential for extending battery life and ensuring safety. This study proposes a novel State of Health estimation model that integrates transfer learning with a Retentive Network architecture, addressing the limitations of traditional methods in data processing and feature extraction. By pre-training a model adaptable to various battery types, the approach leverages the Retentive Network’s powerful temporal data processing capabilities, enabling end-to-end application to raw data and reducing the need for complex preprocessing. The model utilizes a dual-stream parallel network architecture to extract valuable information from both charging and discharging data, significantly enhancing the accuracy of State of Health estimation. Furthermore, the model can quickly adapt to new or sparsely sampled battery types through pre-training and fine-tuning strategies, minimizing reliance on large labeled datasets. Compared with existing transfer learning approaches, the proposed model demonstrates superior performance in terms of accuracy and robustness, particularly in handling long-term degradation patterns. Extensive experiments on multiple public datasets show that the model consistently achieves a Mean Absolute Error below 0.90% and a Root Mean Square Error below 0.95%, confirming its effectiveness and precision. This research provides novel insights and methodologies for optimizing battery performance and improving management practices.