Assessing driving behavior influence on fuel efficiency using machine-learning and drive-cycle simulations
Amin Mohammadnazar, Zulqarnain H. Khattak, Asad J. Khattak
Abstract
Consumption of fossil fuel-based energy for vehicle propulsion and associated emissions are a global concern. One pathway to energy reduction is to examine situations where high-energy consumption occurs on roadways, e.g., speed volatility at work zones, or on sharp curves, which has been understudied. Harnessing second-by-second data from the naturalistic driving study and using the concept of driving volatility, this paper explores driving styles in work-zones and curves using machine learning approaches (k-medoids, hierarchical clustering) and drive-cycle simulations from Autonomie®. Results show that aggressive driving account for 12.2 % and 15.4 % of events that occurred in work zones and on curves and leads to 23 % increase in fuel consumption as opposed to normal driving. These results have implications for transportation agencies to improve work-zone configurations and provide signage or technology on curves to reduce fuel consumption and emissions. Moreover, automated and connected vehicles can smooth out traffic flow with advanced advisories and warnings.