Abstract

Accurate energy consumption prediction and therefore velocity prediction, over mid-term is essential for efficient vehicle operation, particularly in electric and hybrid vehicles where range and resource management are critical. However, most current models fail to account for the real-time variability in driver behaviour, especially in response to dynamic traffic situations.  This work proposes an adaptive system that identifies and adapts to behaviour patterns which are associated with repeatedly inaccurate predictions by comparing expected and actual driver actions. This allows the system to automatically learn from new behaviour patterns without the need for manual supervision. The proposed method is evaluated on exemplary driving situations extracted from open driving data. By fusing traffic context modelling with driver behaviour modelling, this method contributes to accurate velocity prediction systems, which can adapt to changes in driver behaviour over time.