Traffic control algorithms are not always able to efficiently utilize the network capacity causing economical and societal costs. The main complicating factor of network-wide traffic control is (simply) the size of the network, especially when controlling the traffic in an entire urban region — i.e. a densely populated area housing several millions of people. Controlling the traffic in such a region requires the coordination of several hundreds of actuators, such as variable speed limits (VSLs), ramp metering (RM), traffic lights, and route guidance. This is a challenging problem from a computational point of view due to the large amount of decision variables, but also from a theoretical point of view due to the many problem characteristics that need to be accounted for. A promising approach to control the traffic in very large networks is to divide the network into sub-networks. A sub-network is defined in this dissertation as a medium-to-large scale network consisting of tens of kilometres of freeway or tens of intersections. The sub-network controllers are used to optimize the performance in the sub-networks while a higher level controller optimizes the flows that are exchanged between the sub-networks leading to network-wide performance improvement. In this way, the sub-network controllers can consider more detail while the higher level controller can consider more simplified or aggregated dynamics. This dissertation focuses on the design of algorithms for sub-networks in the light of a multi-level or hierarchical system as discussed above. Two types of sub-networks are considered, namely, freeway and urban sub-networks. Ideally, a freeway or urban traffic control algorithm is able to automatically select the control signals that maximize the sub-network throughput in different (traffic) situations. Although various optimization-based algorithms have been proposed to achieve that goal, this type of algorithm has not been implemented in practice due to several reasons, namely; 1) the computational complexity of the optimization problem, 2) the noise and uncertainties involved when estimating traffic states and predicting disturbances, and 3) the not very insightful optimized control actions. In contrast to that, mainly non-optimizing control algorithms of the feedback or the feed-forward type are implemented in practice. Advantages of these algorithms are that they require little computation time, that they do not rely on demand predictions, and that they exploit simple or insightful algorithmic formulations. However, they may not be able to optimize the performance in all traffic situations. Recent technological innovations and scientific insights provide opportunities for improving both freeway and urban traffic control algorithms. Technological innovations, such as the proliferation of in-vehicle technology enabling cooperative systems, can be used to provide better detection and actuation possibilities that may be used to improve the controller performance. Similarly, scientific insights may be used to develop new algorithms that make more efficient use of existing detection and actuation possibilities. In some cases, a combined approach may be followed in which new algorithms are developed that make efficient use of new detection and actuation possibilities. Given the network-wide traffic control problem and the opportunities to improve traffic control algorithms as discussed above, the main aim of this dissertation is the design of computationally efficient traffic control algorithms for throughput improvement of medium-to-large scale freeway or urban traffic networks that: • coordinate the control actions of (different types of) actuators at different locations in the network, • take the impact of the control actions on the network-wide performance over a time horizon into account. The main research objective is achieved by developing several algorithms for the control of traffic in freeway networks (part I) and urban traffic networks (part II) as discussed below. Part I — Freeway traffic control: Cooperative systems can be used to develop more efficient freeway traffic control algorithms when compared to existing, purely infrastructure-based systems. The reason for this is that using in-vehicle technology may provide more accurate and faster detection and actuation possibilities. However, not many approaches for the coordinated control of individual vehicles to control the traffic flows on an entire freeway stretch exist. To this end, Chapter 2 proposes a cooperative speed control algorithm to resolve jam waves in order to improve the freeway throughput. The algorithm — called COSCAL v1 — uses the individual vehicles as detectors and actuators assuming a 100% penetration rate. The road-side system computes based on floating car data (FCD) which driving strategy vehicles on the freeway have to follow between which locations on the freeway to resolve the jam wave and stabilize the traffic. Simulations using microscopic simulation show that the algorithm is able to improve the freeway throughput by resolving a jam wave using a negligible amount of computation time. Hence, this chapter shows that the it is possible to develop efficient algorithms for the control of traffic flows using cooperative systems. The application of control strategies that optimize the flows between different network elements — e.g. on-ramps, off-ramps, bottlenecks, and segments — has the potential to improve the freeway performance as well. One of the main issues of this type of algorithms is balancing the required computation time and performance of the control strategy. Hence, Chapter 3 proposes a computationally efficient model-based predictive control (MPC) strategy for coordinating VSLs and RM installations in order to improve the freeway throughput. The balance between computation time and performance is improved by reducing the number of optimization variables through parameterization of the VSL and RM signal. The parameterized VSL signal consists of the speed with which the downstream and upstream boundaries of a speed-limited area propagate. The parameterized RM signal consists of the density set-points of a feedback RM strategy based on the ALINEA algorithm and the time when the settings of the feedback strategy are changed. The approach is evaluated using macroscopic simulation for two different cases, namely, when resolving a jam wave, and when preventing congestion caused by a high on-ramp demand. It is shown that the proposed MPC approach can realize throughput improvements of 12% and 10% respectively while realizing a better balance between computation time and throughput compared to a non-parameterized MPC strategy. Part II — Urban traffic control: Improving the throughput of urban traffic networks is a complex problem due to, among others, the discontinuous nature of the intersection flows, the large number of actuators, and the characteristics of the urban traffic dynamics. To the best knowledge of the author, a computationally efficient optimization algorithm for the coordination of intersection flows that can realize good performance in all traffic regimes is currently lacking. Therefore, Chapter 4 proposes an efficient linear MPC strategy for optimizing the traffic flows in order to improve the urban road network throughput. The proposed MPC strategy uses the link transmission model (LTM) as the prediction model and aggregates the traffic flow dynamics to tens of seconds. So, instead of green-times, the fractions of green-time used by every stream are the optimization variables, which are real-valued. It is shown using macroscopic simulation that the use of the LTM leads to a better balance between computation time and realized throughput when compared to a linear MPC strategy based on the cell transmission model. It is also shown that the inclusion of upstream propagating waves leads to better throughput when compared to a linear MPC strategy based on the store-and-forward model but also that the MPC strategy requires more computation time. The application of cooperative systems may lead to improved performance of urban traffic control algorithms. However, it may also cause an interaction effect between the chosen intersection control strategy, and the route choice of the road-users. Hence, in order to maximize the network performance, a control strategy has to account for the impact of the control signals onto the route choice and potentially control the route choice itself. However, jointly optimizing the signal timings and route choice is a computational complex problem. Chapter 5 proposes an efficient optimization strategy for the control of flows and routing decisions in order to improve the network throughput. The inclusion of routing decisions results in a non-linear prediction model and optimization problem. Therefore, an efficient optimization algorithm of the sequential linear programming (SLP) type is used and an analytic procedure to approximate the gradient in an operating point is proposed. It is shown using macroscopic simulations that the algorithm can realize a better balance between computation time and throughput when compared to applying a conventional numerical optimization algorithm. The algorithms proposed in Chapter 4 and Chapter 5 both assume that the traffic flows at intersections are continuous. However, intersection flows are discontinuous so that directly optimizing the signal timings leads to a discontinuous optimization problem. Solving such a problem is not feasible in real-time when applied to medium-to-large scale networks. Hence, an alternative approach may be needed that can coordinate the signal timings in a network without directly optimizing the signal timings. To this end, Chapter 6 proposes a hierarchical control framework to coordinate the signal timings in order to improve the urban network throughput. The framework consists of two layers. The top layer uses the MPC strategy proposed in Chapter 4 to optimize the aggregated flows at intersections. The bottom layer consists of the individual intersection controllers which actuate at every time-step the stage that leads to the best tracking of the optimized outflows. Evaluations using macroscopic simulation are carried out to study the added value of the network coordination layer, and the impact of the timing onto the controller performance. Evaluations using microscopic simulation demonstrate the controllers ability to improve the throughput by distributing the queues over the network when compared to maximizing the outflow of the individual intersections without coordination even when subject to a larger mismatch between prediction and process model. In conclusion this dissertation proposed several computationally efficient networkwide traffic control algorithms for throughput improvement of medium-to-large scale freeway or urban traffic networks. These algorithms are designed to coordinate the control actions of (different types of) actuators at different locations in the network and to take the impact of the control actions on the network-wide performance over a time-horizon into account. This is realized by exploiting new features of in-vehicle technology enabling cooperative systems to provide better detection and actuation possibilities and by using recent scientific insights to develop more efficient algorithms. Various directions for further research are proposed. First, additional research is required to integrate the proposed algorithms into a hierarchical of multi-layer framework for the coordinated control of entire urban regions. Second, the algorithms proposed in this dissertation may be further improved, for instance, by further improving the balance between computation time and performance, by further exploiting the potential of in-vehicle technologies, or by studying the impact of relaxing the assumptions used in this dissertation. Third, recommendations for the application of concepts in practice are presented. (Author/publisher)
Abstract