State of charge prediction for lithium-ion batteries based on multi-process scale encoding and adaptive graph convolution

Accurate prediction of the state of charge (SOC) of lithium-ion batteries in spacecraft power systems is crucial for optimizing energy management, effectively responding to emergencies, and enhancing system reliability and safety. To capture the slow degradation of chemical characteristics across multiple charge-discharge processes and cycles in lithium-ion batteries, this paper proposes a SOC prediction method based on multi-process scale encoding and adaptive graph convolution (MPSEAGC). This method employs adaptive graph convolution to capture the similarities and gradual differences between various charge-discharge processes of the lithium-ion battery and uses a multi-head attention mechanism to identify intra-sequence correlations within each charge or discharge process. By conducting comparative experiments using publicly available lithium-ion battery datasets, the proposed method achieved a mean squared error (MSE) of 0.062 % and a mean absolute error (MAE) of 1.652 %, representing improvements of 54.74 % and 42.26 % over existing models, particularly excelling in long-sequence prediction. This method provides an accurate and reliable approach to predicting the SOC of batteries, offering a promising solution for predicting battery performance and state in practical applications.