Abstract
Road transportation has a good share in the global carbon dioxide emissions and models for estimating the vehicular emissions using physical parameters help to understand and potentially reduce them. However, such models at a regional level are usually insensitive to driving dynamics, as they are based on the average speed of the vehicles. The existing models considering the instantaneous speed and acceleration also use other factors like weather, vehicle parameters, etc., which involves many measurements. Furthermore, the estimation is done at large time periods of the order of several seconds. In this work, a real-time time series data is used to develop a model using only the vehicular speed and acceleration. It is based on a novel technique of using windows of the driving dynamics captured in a very short period, making some assumptions. The optimal drive features that influence the fuel consumption have been estimated using many machine learning regression models, validated, and compared. Among them, a multi-layer perceptron resulted the highest cross-validation of 0.64 using only the window of speeds, which is concluded to be reasonably good for practical estimation. Finally, these models are aimed to be applied in real applications based on J-credit and eco-driving.
