A detailed and accurate fuel model fuel consumption model that reflects real-world fuel consumption is required as input for devising and executing a model policy for prospective regulatory tools. The fuel consumption model based on the vehicle-specific power (VSP) has rapidly become the primary development direction since the release of the Motor Vehicle Emissions Simulator (MOVES) model. However, fuel consumption cannot be accurately characterized under high-speed scenarios. This work develops two fuel consumption models for the light-duty (gasoline) vehicles that can better characterize fuel consumption for light-duty vehicles under high-speed scenarios. For model 1, the VSP of −5kW/ton is a crucial turning point. When VSP∈ [−30, −5] kW/ton, the fuel rate is only determined by speed. When VSP∈(−5, 30], the fuel rate will gradually increase with VSP, and the growth characteristics will vary with speed. Model 2 develops the new interpretations for VSP and forms the one-to-one correspondence between the fuel rate and the new VSP. The two models can separately improve the accuracy by 12.2% and 13.8% compared with the conventional model. The fuel factor differences become significant when speed is higher than 65 km/h, which are separately 30.66% and 28.13% higher than the conventional VSP model when the speed is 100 km/h. Further, the fuel factors of the two models for freeways are, respectively, 6.33% and 7.56% higher than the conventional VSP model, and the distinction for arterial, collector, and local street roads is not notable.

The use of vehicle operating mode (OpMode) distribution is widely accepted for estimating energy consumption and emissions in the Motor Vehicle Emission Simulator (MOVES) model. However, the heterogeneity of driving behavior may lead to errors when using the default OpMode distribution. To improve the accuracy of energy consumption estimations, it is necessary to recognize the heterogeneity in OpMode distribution among different driving behaviors. With this aim, this paper designs a speed-specific indicator of energy efficiency reflecting driving behavior based on the speed-specific vehicle-specific power (VSP) distribution. The paper uses field data from 26,082 drivers recorded second by second during workdays. It also discusses the intra-heterogeneity and inter-heterogeneity of driving behavior based on unsupervised algorithm clustering. The findings of this paper are as follows. (1) The speed-specific VSP distribution clearly reflects the differences in energy efficiency of individuals’ driving behavior. (2) The energy efficiency indicator reflects the multidimensional inter-heterogeneity and intra-heterogeneity of driving behavior. (3) Drivers’ varied driving behavior causes heterogeneity in energy efficiency at different speeds, possibly causing an error of 6.34% in the emissions estimations. (4) Drivers of electric vehicles (EVs) and hybrid electric vehicles (HEVs) show more aggressive driving behaviors than drivers of conventional vehicles (CVs), which may cause an energy estimation error of over 6% for EVs and HEVs. Thus, the OpMode distribution of EVs, HEVs, and CVs should be modeled separately for on-road energy estimations.