Capacity Forecasting of Lithium-Ion Batteries Using Empirical Models: Toward Efficient SOH Estimation with Limited Cycle Data

Accurate prediction of lithium-ion battery capacity degradation is crucial for reliable state-of-health estimation and long-term performance assessment in battery management systems. This study presents an empirical modeling approach based on experimental data collected from four lithium iron phosphate (LFP) battery packs cycled over 75 to 100 charge–discharge cycles. Several mathematical models—including linear, quadratic, single-exponential, and double-exponential functions—were evaluated for their predictive accuracy. Among these, the linear and single-exponential models demonstrated strong performance in early-cycle predictions. It was found that using 30 to 40 cycles of data is sufficient for reliable forecasting within a 100-cycle range, reducing the mean absolute error by over 80% compared to using early-cycle data alone. Although these models provide reasonable short-term predictions, they fail to capture the nonlinear degradation behavior observed beyond 80 cycles. To address this, a modified linear model was proposed by introducing an exponentially decaying slope. The modified linear model offers improved long-term prediction accuracy and robustness, particularly when data availability is limited. Capacity forecasts based on only 40 cycles yielded results comparable to those using 100 cycles, demonstrating the model’s efficiency. End-of-life estimates based on the modified linear model align more closely with typical LFP specifications, whereas conventional models tend to underestimate the cycle life. The proposed model offers a practical balance between computational simplicity and predictive accuracy, making it well suited for battery health diagnostics.