Lithium-Ion Battery State-of-Health Estimation Method Based on Hierarchical Convolution and Multichannel Interactive Dense Transformer Network

Accurate state-of-health (SOH) estimation plays an important role in the safe and reliable operation of lithium-ion batteries (LBs). Recently, Transformer has become a hot architecture for SOH end-to-end estimation, driven by its capability in sequence modeling. However, existing methods mine the individual dependencies of detection data (voltage, current, and temperature) from the temporal or spatial (variable) dimension, ignoring the joint dependencies of the spatiotemporal synergistic dimensions, leading to inadequate extraction of aging features. Furthermore, the skip connection mechanism only exists within the encoder, inevitably losing discriminative feature information during forward propagation among encoders. Thus, this article proposes a hierarchical convolution and multichannel interactive dense Transformer network (HCMDT-Net) for SOH estimation. Specifically, this article first designs a convolution-based local encoding enhancement module (LEEM) to embed the local temporal information of monitoring data and enhance spatial representations. Second, temporal channel attention (TCA) is devised to capture the joint dependence of the data from spatiotemporal synergistic dimensions. Meanwhile, to prevent the loss of discriminative feature information, a cross-layer adaptive fusion mechanism (CAFM) is devised to improve information transitivity among encoders. Finally, the accurate estimation of SOH is achieved in combination with a multilayer perceptron (MLP). Extensive experiments are conducted on two benchmark lithium battery aging datasets. Experiment results show that the proposed method performs better than several existing methods.