BatteryMFormer: Multi-Level Transformer for Battery Degradation Forecasting
A new AI model, BatteryMFormer, has been proposed for early battery degradation trajectory forecasting (BDTF). The model addresses two key characteristics of battery degradation data: a multi-level structure with regularities shared within aging conditions and trajectory patterns across batteries, and degradation-related variations in voltage-current profiles localized to specific state-of-charge (SOC) intervals. BatteryMFormer integrates an aging-condition-aware decoder that injects aging-condition priors via queries and attention, and a meta degradation pattern memory that learns and reuses shared patterns. The approach aims to improve prediction of full-life state-of-health trajectories from early operational data, critical for battery optimization, manufacturing, and deployment. The research is detailed in arXiv preprint 2605.27044.
Key facts
- BatteryMFormer is a multi-level Transformer for early battery degradation trajectory forecasting.
- It addresses multi-level structure in degradation data and localized SOC variations.
- The model includes an aging-condition-aware decoder and meta degradation pattern memory.
- Early BDTF predicts full-life state-of-health from early operational data.
- The research is published on arXiv with ID 2605.27044.
- The approach aims to improve battery optimization, manufacturing, and deployment.
- Existing methods fail to explicitly model multi-level and localized characteristics.
- The model injects aging-condition priors via queries and attention.
Entities
Institutions
- arXiv