Lithium-ion battery calendar health prognostics based on knowledge-data-driven attention

In real industrial electronic applications that involve batteries, the inevitable health degradation of batteries would result in both the shorter battery service life and decreased performance. In this paper, an attention-based model is proposed for Li-ion battery calendar health prognostics, i.e., the Capacity Forecaster based on Knowledge-Data-driven Attention (CFKDA), which will be the first work that applies attention mechanism to benefit battery calendar health monitor and management. By taking the battery empirical knowledge as the foundation of its crucial part, i.e., the knowledge-driven attention module, the CFKDA has realized a satisfactory combination of the complementary domain knowledge and data, which has improved both its theoretic strength and prognostic performance significantly. Experimental studies on practical battery calendar ageing demonstrate the superiority of CFKDA in forecasting and generalizing to unwitnessed conditions over both state-of-the-art knowledge-driven and data-driven calendar health prognostic models, implying that the introduction of domain knowledge in CFKDA has brought a significant performance improvement. Moreover, error analysis shows that temperature is a more significant influencing factor than State of Charge (SoC) in terms of calendar degradation mode, which provides the reference value for battery management