Complete charging-curve prediction and critical states estimation of lithium battery based on improved transformer and partial sampling points

Abstract

Energy storage (ES) is regarded as a key enabler to decarbonize power systems. Accurate state estimation of battery energy storage systems is crucial for efficient battery utilization and prolonging battery life. However, it is often hindered by fragmented data caused by uncertainties in the charging/discharging start points. In this paper, we propose a novel method for predicting the complete charging curve and estimating the critical states of lithium batteries by utilizing partial sampling data. A multi-scale interval attention (MSIA) mechanism is introduced to capture information at different granularities from the charging curve. Transformer model based on MSIA enables us to predict the complete charging curve and gain richer information compared with the traditional state-of-charge prediction and state-of-health estimation methods. Experimental results with multiple datasets demonstrate that the proposed method excels in predicting the complete charging curve and estimating states of lithium batteries.