Fuel Consumption Modeling and Condition-Corrected Benchmarking of Heavy-Duty Trucks Using Longitudinal OBD Data

Fuel-efficiency comparisons among heavy-duty trucks are confounded by heterogeneous operating conditions, limiting fair assessment of vehicle-level energy performance. This study develops a longitudinal on-board diagnostics (OBD)-based framework for trip-level fuel consumption modeling and condition-corrected benchmarking. The dataset comprised 104,698,864 second-level records from 60 in-use trucks. After temporal reconstruction, plausibility screening, and trip extraction, 23,883 valid trips covering 596,627.70 km were retained, of which 23,770 were used for predictive modeling. Motion, engine operation, thermal state, and speed–load coupling features were evaluated using five-fold vehicle-grouped cross-validation. HistGradientBoosting achieved an out-of-fold R2 of 0.936, an RMSE of 6.97 L/100 km, and a MAPE of 11.98%. High- and very-high-load operation accounted for only 8.59% of operating time but 37.40% of fuel use, highlighting the disproportionate energy demand of high-load operation. Vehicle-level out-of-fold residuals were then combined with operating exposure to support condition-corrected benchmarking. The framework provides a practical basis for fuel-efficiency screening and fleet energy management prioritization without interpreting residual deviations as direct evidence of mechanical faults.