Integrated Control of Mixed Traffic Networks using Model Predictive Control

Motivation The growth of our road infrastructure cannot keep up with the growing mobility of people, and the corresponding increase in traffic demand. This results in daily congestion on the freeways. It is an illusion that the problem of congestion can be solved completely within a few years, but it is possible to improve the current situation. This is necessary since the congestion on the roads has disadvantages for the drivers, including long travel times and high economic costs. It has also disadvantages for the surroundings of the roads, where the increased traffic load results in e.g., pollution, noise, and unsafety in residential areas. The goal of this thesis is to develop model-based traffic control methods that improve the situation for the drivers as well as for the environment. Mixed urban and freeway control Due to the growing density of the road networks, freeways and urban networks become tightly coupled. This requires that the control on the two types of roads should also be coupled. Therefore we develop a mixed urban-freeway model that combines a macroscopic freeway model with an urban queue length model. For the macroscopic model we use the traffic flow model METANET. The urban queue length model is based on a model developed by Kashani and Saridis, extended with horizontal queues, blocking effects, and a shorter time step. The two models are coupled via the modeling of on-ramps and off-ramps. The obtained macroscopic model can simulate traffic flows efficiently, and thus is suitable for the use in a model-based control setting. We develop such a model-based control method that uses model predictive control, with the mixed urban-freeway model as prediction model. Route choice control Control measures can also be used to influence route choice. Route choice is a complicated process that can be divided into two main processes with a different time scale. The withinday route choice focuses on the choices that drivers make during their trip, while the dayto- day route choice describes the change in route choice from one day to the next. We first discuss the effect of ramp metering on within-day route choice. By installing a ramp metering installation at an on-ramp, the density – and thus the travel time – on the freeway as well as on the on-ramp itself is changed, which influences the route choice. We develop two different methods to include route choice in model-based controllers: a dynamic traffic assignment model, and a model based on a look-up table determined via Bayesian learning. Second, we investigate day-to-day route choice using the Bayesian learning model. We assume that drivers base their route choice on a combination of the density and the corresponding travel times experienced on previous days. With the model based on Bayesian learning, in addition to the on-ramp metering, we also briefly explore the effects that can be obtained with the use of off-ramp metering. Another measure that can be used to influence the route choice is displaying travel time information on dynamic route information panels. Displaying travel times with a large enough difference can encourage drivers to change their route choice. We model the drivers’ reaction on the route information, and develop a controller that actively influences the route choice of the drivers using information on dynamic route information panels in combination with variable speed limits. Since route choice models as described above in general require large computational efforts, we also formulate a simplified route choice model for day-to-day route choice that can be used to obtain fast predictions of the route choice behavior, and that is suitable to obtain a first impression of the traffic assignment, for use in on-line optimization algorithms, or as initial value for more complex optimization algorithms. We use this model in a modelbased control setting where the objective of the controller is to influence the route choice, and investigate in particular speed limit control and outflow control. Practical control issues To apply model based controllers in practice, several practical issues have to be considered. We present a short overview of interesting issues, and next we explicitly investigate the effect of averaging method that is used for the speed measurements. We compare the time mean speed, harmonic mean speed, geometric mean speed, time average space mean speed, and the estimated space mean speed based on instantaneous speed variance and based on local speed variance. All averaging methods are applied in a freeway speed limit control method, to investigate the influence of the averaging method on the controller performance. Conclusions The current traffic infrastructure can be used more effectively when advanced control algorithms are used. Existing traffic control measures can be used to decrease the costs for the drivers, and to relocate the traffic flows via influencing the route choice. This results in economical benefits due to shorter travel times, environmental benefits due to the reduction of pollution and noise, and safety benefits due to the lower flows in urban areas