Battery aging inconsistency prediction immediately after manufacturing with formation data

The formation process of power batteries is recognized as a crucial factor influencing lifespan and safety. Here, a prediction method for aging inconsistency based on formation data is proposed, where the cumulative time within different voltage intervals during the formation period is extracted as features from 30 lithium-ion batteries and 39 sodium-ion batteries, followed by feature screening through Pearson correlation analysis. Gaussian Process Regression prediction models are constructed using squared exponential kernel, rational quadratic kernel, Matern3/2 kernel, and Matern5/2 kernel respectively for aging inconsistency prediction. Prior to prediction, Z-score standardization is applied to sodium-ion batteries due to significant lifespan deviations in some cells. For lithium-ion batteries, the mean absolute percentage error of capacity prediction under baseline formation conditions remains below 1.5%, while fast formation conditions achieve MAPE below 0.3%. Both conditions demonstrate MAPE values below 6% for predictions of lithium loss, positive electrode active material loss, and negative electrode active material loss. Sodium-ion batteries exhibit prediction MAPE of 5.23% under pulse formation conditions and 2.64% under C/20 formation conditions. This work overcomes the temporal limitations of conventional aging tests by enabling immediate prediction upon production completion. Meanwhile, it establishes a scientific foundation for battery production optimization and consistency grouping.