Capture long-run seasonality in forklift battery SoH time series

Determine whether the state-of-health (SoH) time series of lithium-ion battery packs used in electric forklifts exhibit long-run seasonality and develop predictive modeling approaches that accurately capture such seasonal behavior in these SoH trajectories.

Background

The paper develops and evaluates machine learning models—particularly gradient boosting—to forecast the state-of-health (SoH) time series for 45 lithium-ion battery packs used in electric forklifts. The data were aggregated to regular daily time series, and various ensemble and neural network methods were compared, with the best model achieving low error metrics over short horizons.

Despite competitive performance, the authors note limitations related to the available time span (approximately 32 months) and potential long-run dynamics, specifically the possibility of seasonal effects in the battery SoH that their models did not capture. Addressing seasonality is important for improving reliability of long-term forecasts and maintenance planning for industrial fleets.