A Transformer Encoder for Real-Time Health Diagnosis and Aging Modeling of Lithium-Ion Batteries

July 24, 2026

Accurate state-of-health (SOH) estimation is crucial for battery aging management, yet raw operational data often contains high-frequency measurement noise and volatile fluctuations, which degrade the precision of health indicator extraction. A hybrid denoising and deep-learning framework is therefore proposed: a joint Savitzky–Golay (SG) filter and Kalman filter (KF) scheme is first deployed to smooth charging profiles and eliminate multi-scale noise, from which clean incremental capacity (IC) curves are reconstructed to robustly extract key aging features. These optimized health indicators are subsequently fed into a Transformer model, leveraging its self-attention mechanism to capture long-term degradation trends for real-time SOH prediction. Experimental validations demonstrate high tracking accuracy with significantly reduced root-mean-square errors and robust generalization. The algorithm effectively balances noise immunity with deep feature extraction, making it highly suitable for online BMS health prognostics.

Authors in the community:

• 

  H

  HanFei Gu

  ist1115704

Supervisors of this institution:

• 

  A

  António José Castelo Branco Rodrigues

  ist12288

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

por - Portuguese

Instituto Superior Técnico