Vehicular fuel consumption estimation using real-world measures through cascaded machine learning modeling

Abstract

Many of the existing studies on vehicular fuel consumption estimation are criticized in aspects such as ignoring real-world training data, low diversity of test fleet, impracticality of models in real-world applications (i.e. instrument-independent eco-driving), or their prediction power in the non-linear multi-dimensional space of fuel consumption estimation. In this paper, we proposed a machine learning modeling method using large on-road data collected from a fleet of 27 vehicles. The usability of models in absence of specialized instruments was in focus. We tried to improve the accuracy of our base models by introducing engine-speed estimates through a cascaded modeling procedure. As a result, the accuracy of models reached 83%, while improvements as high as 37% were achieved depending on the technique (support vector regression or artificial neural networks) and vehicle class. Finally, we took the first step from vehicle-specific models towards category-specific modeling by a categorical analysis over fleet attributes.