Optimizing Cycle Life Prediction of Lithium-ion Batteries via a Physics-Informed Model

TL;DR

This work tackles Li-ion battery cycle-life prediction by fusing a physics-based Arrhenius-law–inspired capacity-loss model with a self-attention regressor that infers its parameters from early-cycle data. The physics component uses $\hat{Q}_{loss}(x) = e^A x^B + C$ and derives life as $\ell = [e^{-A}(0.2 - C)]^{1/B}$, achieving $R^2$ up to $0.994$ on the Severson dataset. A self-attention module predicts $(\hat{A}, \hat{B})$ from a carefully selected set of early-cycle features, enabling reconstruction of full capacity-loss curves and threshold-agnostic end-of-life predictions. Two-stage training yields competitive cycle-life RMSEs (primary $=127.83$ cycles, secondary $=179.92$ cycles), offering robustness and interpretability while preserving the ability to redefine end-of-life thresholds via the reconstructed curves.

Abstract

Accurately measuring the cycle lifetime of commercial lithium-ion batteries is crucial for performance and technology development. We introduce a novel hybrid approach combining a physics-based equation with a self-attention model to predict the cycle lifetimes of commercial lithium iron phosphate graphite cells via early-cycle data. After fitting capacity loss curves to this physics-based equation, we then use a self-attention layer to reconstruct entire battery capacity loss curves. Our model exhibits comparable performances to existing models while predicting more information: the entire capacity loss curve instead of cycle life. This provides more robustness and interpretability: our model does not need to be retrained for a different notion of end-of-life and is backed by physical intuition.

Figures (9)

• Figure 1: (a) Discharged capacity, $Q_d(t)$, and voltage, $V(t)$, measured for one battery over the course of a charge-discharge cycle. The discharge portion of the cycle is boxed in black. (b) The discharge-voltage curve, $Q_d(V)$, for the same battery. Note that the x-axis in (b) is flipped to reflect voltage decreasing over the course of the discharging process.
• Figure 2: (a, b) Evolution of the discharge-voltage curve over cycles for two batteries with different cycle lives. Curves from cycles evenly spaced between 1 and 100 are plotted and distinguished by saturation. As cycle number increases, the curve progressively sags more for the battery with lower lifetime. (c, d) $\Delta Q_{100-10}(V)$ for the same two batteries. (e) $\Delta Q_{100-10}(V)$ plotted for all batteries in the dataset, with shade corresponding to cycle life.
• Figure 3: (a-c) Capacity loss model fit to three capacity loss curves using least-squares. The three curves reflect batteries with substantially different lifetimes, demonstrating the ability of the model to generalize. $R^2$ for each individual battery is displayed, and average $R^2$ across all 124 batteries in the dataset is 0.976. (d) Cycle lives derived from the fitted capacity loss curves, plotted against true cycle lives. We observe $R^2=0.994$, demonstrating high goodness of fit.
• Figure 4: Schematic of the physics-based model. One half utilizes an Arrhenius Law-inspired model to capture capacity loss curves. The other half utilizes a self-attention layer to predict Arrhenius Law parameters from early-cycle data.
• Figure 5: Feature correlation scores of the five features most correlated with $\hat{A}$ and $\hat{B}$.