Development of a driving resistance prediction model for light-duty vehicles using large-scale coast-down test data

초록

Speed–Vehicle Specific Power (VSP)-based real-world CO2 estimation can improve accuracy compared with a speed-based model. However, VSP requires vehicle-specific road-load coefficients ((Formula presented), (Formula presented), (Formula presented) ) that are difficult to measure, so studies often use approximations or representative values. This study developed practical, specification-based models to estimate road-load coefficients for light-duty vehicles (LDVs) and evaluated their suitability for speed-VSP based real-world CO2 estimation. Coast-down data from 122 gasoline and diesel LDVs were used, and 12 diesel vehicles were held out for external validation. Stepwise regression was applied using curb weight, overall length, width, height, tire width, and tire radius as candidate predictors. The final models selected curb weight, overall height, and overall length for (Formula presented), while (Formula presented) retained overall height only; (Formula presented) was explained by overall height and tire radius. External validation was conducted using real driving emission data from the 12 vehicles, consisting of 7 SUVs and 5 non-SUVs. VSP distributions computed using the predicted coefficients closely matched those computed using measured coefficients, and trip-level CO2 estimates based on predicted-coefficient VSP were comparable to those based on measured-coefficient VSP. Compared with a speed-only approach, the speed–VSP approach substantially reduced errors: mean absolute error decreased from 18.9% to 10.1% for SUVs (RMSE: 44.1 to 24.3 g/km) and from 18.3% to 9.0% for non-SUVs (RMSE: 43.2 to 19.1 g/km). Overall, these results indicated that road-load coefficients could be approximated from basic specifications without coast-down testing and still support practical VSP-based real-world CO2 evaluation.

키워드