On-device adaptive battery power prediction reduces error by up to 14.88%

Researchers introduce a method for on-device learning that allows electric vehicles to continuously adapt pretrained battery power prediction models to new, unseen data distributions. This approach transforms existing models into adaptable versions that retain critical hyperparameter knowledge from initial training while enabling continuous updates in resource-constrained systems.

• The study investigates both online and offline model adaptation strategies for EVs.
• Online adaptation techniques achieve mean absolute error reductions of up to 7.49%.
• Offline adaptation techniques achieve mean absolute error reductions of up to 14.88%.
• Results show significant improvements in forecasting performance across various models and time horizons.

This study highlights the substantial benefit of on-device adaptation, resulting in enhanced battery power predictions compared to unadapted model deployments in real-world EV scenarios.