How to Model Real-World Driving Behavior? Probability-Based Driver Model for Energy Analyses

A wide variety of applications such as driver assistant and energy management systems are researched and developed in virtual test environments. The safe testing of the applications in early stages is based on parameterizable and reproducible simulations of different driving scenarios. One possibility is modeling the microscopic driving behavior to simulate the longitudinal vehicle dynamics of individual vehicles. The currently used driver models are characterized by a conflict regarding comprehensibility, accuracy and calibration effort. Due to the importance for further analyses this conflict of interests is addressed by the presentation of a new microscopic driver model in this paper. The proposed driver model stores measured driving behaviors with its statistical distributions in maps. Thereby, the driving task is divided into free flow, braking in front of stops and following vehicles ahead. This makes it possible to display the driving behavior in its entirety. The comprehensibility of this driver model is given by its simplicity and the calibration effort is low with existing measurement data. These data are recorded with a testing vehicle by a map- and sensor-based monitoring of the environment and the measurement of internal parameters. The performance of the model is evaluated with these measurements and two other state-of-the-art models. The analysis of the simulation results reveals significant improvements of the presented driver model regarding the mentioned conflict. The new driver model shows the desired suitability for energetic analyses in virtual test environments, which are partly already performed for the optimization of the energy management system of plug-in hybrid electric vehicles.