Estimating Light-Duty Vehicle Emission Factors using Random Forest Regression Model with Pavement Roughness

Abstract

Pavement roughness would affect the running of vehicle movement, and thus possibly impact fuel consumption and vehicle emissions, the numerical relationships and analytical steps of which are, however, not yet well studied. The major objective of this paper is to quantify vehicular emission factors-hydrocarbons (HC), carbon monoxide (CO), oxides of nitrogen (NOx), and carbon dioxide (CO2)-and fuel consumption as a function of pavement roughness (the International Roughness Index [IRI]) and other factors. Within each operating mode identification (OMID) bins of vehicle operational status, a random forest regression model (RFRM) was identified to estimate emission factors and fuel consumption. The field test data, with a total length of 1,067.41 mi driving and 323,075 data pairs from one test vehicle, were used to train and validate models. The portable emissions measurement system (PEMS) and a smartphone application for IRI were employed for the tests in Texas, U.S., roadways. Results show that the optimum roughness conditions for lower emissions and fuel consumption are in categories B and C with moderate roughness. The root-mean-square error (RMSE) during training, testing, and validation processes of the RFRM are within 6.4%, implying a good fit of resulted models. IRI has the most OMID bins as number one predictor, followed by vehicle specific power (VSP) and speed. Through separated modeling for each OMID, the impacts of IRI are successfully grasped. It is recommended conducting more field measurements with more vehicle types. This would help with possible incorporation of vehicle emissions, fuel consumption, and other environmental factors into the pavement design, maintenance, and retrofitting process.