Pyramidal Cascaded Merging Attention Network for Lithium-ion Batteries SOH Estimation

With the rapid development of electric vehicles (EVs), the safety and reliability of lithium-ion batteries (LiBs) have drawn more and more attention. However, existing methods are incompetent for precise state of health (SOH) estimation of such components because of lacking micromesh feature extraction and efficient information utilization. Therefore, a pyramidal cascaded merging attention network (PCMAN) is proposed to accurately estimate SOH of LiBs in this paper. PCMAN is designed with a multi-layers pyramidal cascaded architecture for local and global feature excavation. In each layer, an attention block is adopted to extract health state feature and a merging block is constructed for information aggregation. To demonstrate superiority of the proposed method, experiments are conducted on real-world EVs datasets which are collected from existing battery management system. Comparing with traditional methods, PCMAN achieves remarkable performance with average root mean square error of 2.60% and mean absolute percentage error of 2.27%.