Analysis of individual driver velocity prediction using data-driven driver models with environmental features

Driver behavior is a key factor in vehicle energy demand calculation required for vehicle optimization strategies and range calculations of electric vehicles. Thereby the vehicle speed resulting from the driver demand has a major influence on the energy consumption. Thus, if the individual speed profile of a driver can be predicted accurately, the energy demand of a vehicle for a given route can be estimated, enabling better range calculation and optimization strategies. However, the driver behavior depends on multidimensional input factors, varies from driver to driver and can change over time. In this paper a learning approach is proposed to predict the individual speed profile. This approach takes environmental influences on the driver behavior into account. Different artificial neural network models and a Kernel regression approach for driver velocity prediction are investigated. The learned models are evaluated on real drive data from different drivers on a specified route. Results show that Long Short-term Memory networks can predict the driver behavior very accurately, leading to small prediction error.