Transient fuel consumption prediction for heavy-duty trucks using on-road measurements

Abstract

To confront the lack of robust and accessible models predicting diesel truck fuel consumption, this study develops an Engine-based Correction Model (ECM) using a sample containing vehicle operation and road data from heavy-duty diesel trucks (HDTs) along an urban arterial in Sichuan Province, China. This model is compared with two popular methods, the VT-Micro model and the CMEM model, in terms of goodness-of-fit and out-of-sample prediction performance. The results show that the proposed ECM models explain on average 89.7% of the variation in fuel consumption across the four modes and produce the smallest error, measured by MAPE and RMSE, for predicting each mode. Specifically, the MAPE values of the new ECM models are 7.95% to 25.59% lower than those of the VT-Micro models and 5.74% to 13.27% lower than the CMEM models. The improvement of RMSE values ranges from 0.1780 to 2.0940 cc/s and from 0.0171 to 0.7259 cc/s, compared to the VT-Micro model and CMEM model, respectively. In sum, the new ECM outperforms the other two existing models in both goodness-of-fit and predictive power across all modes. The enhanced performance of ECM can be attributed to the correction component and a model-building process powered by the simulated annealing (SA) algorithm that reliably and quickly weed out ineffective model forms. The results can inform driver training to foster energy-efficient driving behavior and assess the impact on fuel consumption of roadway design policies.