Enhancing Battery RUL Prediction: Regeneration Modeling & Domain Robustness

Reliable remaining useful life (RUL) prediction of lithium‑ion batteries underpins proactive maintenance and lifecycle optimization. Two pervasive issues compromise prediction fidelity: (1) non‑monotonic capacity regeneration events, and (2) domain heterogeneity caused by batch‑to‑batch variations. This talk focuses on our dual contributions: a monotone decomposition technique that segregates the capacity signal into a monotonically decreasing component and a regeneration term, each forecasted via Gaussian processes and deep autoregression for uncertainty‑aware RUL estimates; and a robust transfer‑learning ensemble that leverages early‑cycle kernel regression with domain‑distance–based weighting and transfer component analysis for cross‑batch alignment. Results on multiple datasets demonstrate the efficacy of our approaches in real‑world scenarios.