BatteryMFormer: Multi-level Learning for Battery Degradation Trajectory Forecasting

Researchers introduced BatteryMFormer, a Transformer-based framework for forecasting full battery degradation trajectories from limited early-life data.

The model combines an aging-condition-aware decoder, a prototype memory module, and a dual-view encoder to capture degradation patterns more effectively.

It outperforms baseline methods across four battery domains, showing stronger accuracy and generalization.

Better early-life degradation forecasts could support battery design, manufacturing, deployment planning, and long-term reliability.