Estimating truck fuel consumption with machine learning using telematics, topology and weather data

Abstract

When purchasing new trucks it is crucial to have a good estimate of the future fuel consumption of the different truck models to choose from. The fuel consumption is a function of various external variables like the total weight, the average speed and its variation, topography, weather conditions and others. As opposed to classic approaches using physical models, the research presented in this paper is using machine learning techniques to predict fuel consumption based on independent variables. Most of the used data is provided by telematics systems from more than 500 trucks over two years. Three different ML models are compared and gradient boost proves to be the best method for the scope of the investigated truck categories. Besides obvious significance of features like total weight and topography, the results confirm that available data on weather conditions also help to improve the forecast. The obtained quality of the fuel prediction, measured as rooted mean squared error, outperforms results reported in prior research on this topic.