A multi time-scale deep learning model for lithium-ion battery health perception using soft parameter sharing mechanism

Efficient assessment of battery degradation is paramount for effectively utilizing and maintaining battery management systems (BMSs). This paper introduces an innovative residual convolutional network-gated recurrent unit (RCN-GRU) model to accurately perceive the lithium-ion batteries' health with a multi time-scale, that employs a soft parameter sharing mechanism to identify both the short and long-term degradation patterns. The continuous looped Q(V), T(V), dQ/dV, and dT/dV are extracted to form a 4-channel image, where RCN can automatically extract the features from such an image and the GRU is used to capture the temporal features. By designing a soft parameter sharing mechanism, the model can seamlessly predict the capacity and remain useful life (RUL) in a dual time scale. The proposed method is validated on a large MIT-Stanford dataset comprising 124 cells, where a high accuracy with 0.00477 mean absolute errors (MAEs) for capacity and 83 for RUL. Furthermore, an exploration of the partial voltage fragment reveals the promising performance of the proposed method across various voltage ranges, where the partial voltage segment 2.8∼3.2V receives 0.0107 root mean square errors (RMSEs) for capacity and 140 for RUL.