State of health forecasting of Lithium-ion batteries operated in a battery electric vehicle fleet

Most existing methods for battery state of health (SOH) forecasting have been applied to battery cell data from laboratory operation for training and testing. This work goes beyond that by using battery pack data from real -world vehicle operation. Our data source is a fleet of 550 battery electric vehicles (BEVs). We aim to provide different feature sets that are accessible to the user groups of the SOH forecasting model like private BEV owners, BEV fleet managers, and battery designers. To this end, we investigate histogram-based features and accessible features. Our results show that a state-of-the-art SOH forecasting method based on histogram features works not only on battery cell data from laboratory operation, but also on battery system data from real-world BEV fleet operation. The model was able to learn the dependence of the SOH from the battery load, i.e., BEV usage. Switching from accessible features to the histogram-based features showed an improvement in model perfor-mance of up to 6.1 %. Two use cases for different operating strategies exemplary illustrate how the SOH fore-casting model can be applied.

Automated summary

This study surpasses existing battery state of health (SOH) forecasting methods by using battery pack data from real-world vehicle operation. The results show that a state-of-the-art SOH forecasting method based on histogram features works not only on laboratory battery cell data, but also on real-world battery system data.