A novel state of health estimation method for lithium-ion battery pack based on cross generative adversarial networks

With the increase in battery pack scale and strict requirements for weight and volume, full deployment of sensors is costly and difficult to achieve, and sensor fault may lead to data loss. Accurately assessing the state of health (SOH) with incomplete data poses a significant challenge. To meet this gap, a novel SOH estimation method for battery pack based on cross generative adversarial network (CrGAN) was proposed. Firstly, an adaptive boosting algorithm was introduced to establish the SOH estimation model of cell by integrating extreme learning machine (ELM) and kernel ELM. Then, based on incomplete battery data, a CrGAN was first proposed for data augmentation of all cells in battery pack. This model was designed by introducing cross-attention mechanism and latent space coding to fuse the shape and time-dependent characteristics. Finally, the SOH and inconsistency of battery pack were analyzed, followed by the comparative verification with the existing typical methods. The results show that the proposed method can accurately generate data for the battery pack SOH estimation. Moreover, the comparative analysis of different incomplete data provides suggestions for sensor design of large-scale battery packs. In this case, 10 sensors are the optimal solution with mean absolute percentage error of 0.43 %.