Cured Memory RUL Prediction of Solid-State Batteries Combined Progressive-Topologia Fusion Health Indicators

Abstract

Reliable remaining useful life (RUL) prediction provides a reference for the secure operation of solid-state batteries (SSBs). However, the intricate potential relations of degradation mechanism and limited degradation data in SSBs bring tremendous challenge. Thus, a novel RUL prediction method named cured memory strategic term moments network with attention of degradation information combined progressive-topologia fusion health indicators (IDPHIs-CMSTM) is proposed for SSBs. It is designed to obtain the implicit relations from Euclidean space and non-Euclidean space and increase the predicted precision on limited degradation data while interpretability is guaranteed. Specifically, in IDPHIs, layer-by-layer progressive fusion method with back-connections assisted by learnable dot product attention mechanism is proposed to gain deep fusion degradation health indicators (HIs) in Euclidean space. It mitigates the risk on information loss during the deep fusion HIs construction process for SSBs. Besides, the topological relations of degradation HIs are presented by graph attention network with two-layers (GAT). The IDPHIs are fed into the developed novel CMSTM to realize RUL prediction. The motivation for CMSTM comes from the early phases of the Ebbinghaus forgetting process, in which recent historical information is utilized to mitigate the forgetting rate of recent degradation information while exploring implicit relations of different degradation information in limited samples. Experiment results on real SSBs dataset show that the IDPHIs-CMSTM achieves higher than 95% predicted precision with well interpretability.