Data driven analysis of usage and driving parameters that affect fuel consumption of heavy vehicles

The aim of this thesis is to develop data mining models able to identify andclassify usage and driving parameters that affect fuel consumption of heavyvehicles. Heavy vehicles open up for a new range of systems aimed atimproving the efficiency and intelligence of road transports. The mostimportant feature of all these systems and services is the fuel efficiency of these vehicles. For developing energy optimized autonomous vehicles and for helping drivers in eco-driving training, there is a need to understand the usage parameters of these vehicles. One part of this is to understand the factors that affect fuel consumption. In this thesis, comparison of usage and driving patternparameters has been done to analyze fuel consumption of heavy vehicles. Theimportance of the parameters has been evaluated depending on its contributionin predicting fuel consumption of heavy vehicles. This particular idea is of huge interest for the company and it will be used in building optimal control strategies for fuel efficiency. Data mining techniques random forest and gradient boosting were used for theanalyses because of their good predictive power and simplicity. Drivingparameters like speed, distance with cruise control, distance with trailer, maximum speed and coasting were found to be important factors that affect fuel consumption of heavy vehicles. Evaluation of performance of each ofthese models based on Nash- Sutcliffe measure demonstrated that randomforest (with an accuracy of 0.808) could do a better prediction of the fuel consumption using the input variables compared to gradient boosting method(accuracy=0.698). The results proved that it is wise to rely on predictiveefficiency of random forest model for future analysis