Modelling Effects of Micro-level Behavior of Connected, Automated Vehicles on Greenhouse Gas Emissions

This study investigates the impact of the driving behavior of connected, automated vehicles (CAVs) on greenhouse gas (GHG) emissions on different road classifications compared to driver-operated vehicles (DOVs). Four road networks were selected for traffic microsimulation and modelling of GHG emissions. The four road networks cover different general road classifications and characteristics, namely freeway, major arterials, major arterial in rural setting, and major arterial with short-spaced intersections. Employing simulations encompassing three driving behaviors (cautious, normal, and aggressive) and varying CAV penetrations from 0 to 100%, the research replicates the forecasted morning peak hour traffic demand in the City of Ottawa in 2031 while introducing fluctuations of ±20% from the predicted demand volume. Combining the variations in traffic demand, CAV behavior, and CAV penetration rate, 39 simulation scenarios were analyzed on each road network. The vehicle trajectories produced by VISSIM microsimulation were used as input file in MOVES software to estimate the GHG emissions while accounting for the micro-level differences in vehicle behavior. The results indicate that cautious CAVs would generally cause a deterioration in traffic performance and increased GHG emissions, while the general trends for normal and aggressive CAVs indicate improved traffic performance and reduced GHG emissions. However, some trend-breaking areas were observed. Regression analysis was performed to model the expected emissions (measured in terms of equivalent CO2 emissions per vehicle-kilometers travelled) as a function of the road, vehicle, and traffic parameters. Models were developed for the four road classifications and had high coefficients of determination (\({R}^{2}\)) and low root mean square error (RMSE) values. The scatter plots on a testing subset, which was not used in the model development, confirmed the goodness of fit of all developed models. Machine learning was also explored as a modelling alternative and showed a great potential for modelling GHG emissions based on the same model parameters.