Prediction and Simulation Models for Intelligent Vehicle Highway Systems

Congestion on highways is becoming a bigger problem every day. With the increasing need for more highway capacity, which cannot always be met by building new highways, opportunities must be found in state-of-the-art technologies for more efficient traffic management. In this thesis traffic control is studied where the traffic consists solely out of intelligent vehicles arranged in platoons, i.e. vehicles that are clustered longitudinally on the highway. With the platooning concept more vehicles can be accommodated on the highway, thus increasing the traffic flow. As of today, no highway simulator is capable of simulating intelligent vehicles. In this thesis, first two existing open-source software packages, MITSIM and FreeSim are assessed regarding their potential for use as a traffic simulator for intelligent vehicles and platoons. As a result of this assessment, FreeSim is selected for further development. FreeSim is transformed such that it is possible to simulate intelligent vehicles. For this, platoon models are implemented. The platoon leaders are equipped with an intelligent speed adaptation (ISA) controller, such that they can follow a reference speed. The followers inside the platoon all have adaptive cruise control (ACC) installed, which makes it possible for them to follow a predecessor vehicle with a safe distance. Since the reference speed for the platoon leaders is not natively provided by FreeSim, FreeSim is adjusted so that a reference speed can be provided in two ways. If the traffic is controlled, the speeds are provided by a controller. For uncontrolled traffic, a METANET model is implemented to compute a reference speed for the platoon leaders. In this thesis we use a so-called “Big Car model” implemented in Matlab as a prediction model. In the Big Car model, a platoon is modeled as one big car with a variable length and has a platoon following behavior. Because FreeSim and the Big Car model use different models, the parameters of the Big Car following model are calibrated. This is done using a nonlinear least squares optimization algorithm. The optimization resulted in a set of parameters that gave a weighted root mean squared error as small as 1.16.