Single-Cycle Identification of State-of-Charge Windows in Large-Format LFP Cells

The state-of-charge (SoC) window used during cycling influences battery behaviour and is relevant to the assessment of cells considered for second-life use. This study investigates whether the SoC window in which a cell was cycled can be identified from the electro-thermal response of a single reference cycle that is identical for every cell, without cycling data recorded within that window. Twenty commercial 102 Ah prismatic lithium iron phosphate (LFP) cells were cycled in five fixed SoC windows with periodic reference cycles. Sixty-five features describing capacity, efficiencies, rest voltages, the voltage curve and surface temperatures were derived from each reference cycle. Under leave-one-cell-out cross-validation, a regularised logistic-regression classifier achieved a per-cycle balanced accuracy of 0.75, and aggregating each cell’s reference cycles correctly classified 19 of 20 cells. With feature-group, model and hyper-parameter selection nested within each fold, the per-cycle balanced accuracy was 0.61 and 16 cells were correctly classified. Misclassifications were concentrated between closely related windows, with a quadratically weighted κ of 0.88 and a depth-of-discharge mean absolute error of 2.0 percentage points. Rest voltages after the full discharge provided the strongest discrimination. The results establish single-cycle window identification for the tested cells and conditions, providing a basis for second-life assessment.