The aim of this master's thesis is to develop a methodology for identifying inefficient segments of a drive and generating feedback that is understandable to the driver, with the goal of improving fuel efficiency in truck driving simulator. Since instantaneous fuel consumption depends not only on driving behaviour but also on road and traffic conditions, the methodology is based on a localized assessment of consumption. A reference consumption level is determined from comparable driving segments, and the deviation from this baseline serves as the target variable for a machine learning model. The methodology was developed and evaluated using data from professional drivers obtained in a truck driving simulator. A LightGBM model was employed to predict consumption deviations, and the SHAP method was used to interpret the model's predictions. A Large Language Model (LLM) was utilized to translate these interpretations into feedback that is easily understood by drivers. The results demonstrate that the developed methodology can successfully predict inefficient driving segments and identify the key factors contributing to increased fuel consumption. This methodology provides a foundation for systems that offer drivers understandable, context-specific, and explainable feedback to enhance driving energy efficiency.
|