MANFIS Based Modeling and Prediction of the Driver-Vehicle Unit Behavior in Overtaking Scenarios

Overtaking a slow lead vehicle is a complex maneuver because of the variety of overtaking conditions and driver behavior. In this study, two novel prediction models for overtaking behavior are proposed. These models are derived based on multi-input multi-output adaptive neuro-fuzzy inference system (MANFIS). They are validated at microscopic level and are able to simulate and predict the future behavior of the overtaking vehicle in real traffic flow. In these models, the kinematic features of Driver-Vehicle-Units (DVUs) such as distance, velocity, and acceleration are used. Unlike the previous models, where some variables of the two involved vehicles are considered to be constant, in this paper, instantaneous values of the variables are considered. The first model predicts the future value of the longitudinal acceleration and the movement angle of the overtaking vehicle. The other model predicts the overtaking trajectory for the overtaking vehicle. The second model is designed for two different vehicle classes: motorcycles and autos. Also, the result of the trajectory prediction model is compared with the result of other models. This comparison provides a better chance to analyze the performance of this model. Using the field data, the outputs of the MANFIS models are validated and compared with the real traffic dataset. The simulation results show that these two MANFIS models have a very close compatibility with the field data and reflect the situation of the traffic flow in a more realistic way. These models can be used for all types of drivers and vehicles and also in other roads and are not limited to certain types of situations. The proposed models can be employed in ITS applications and the like.