Gedecentraliseerd netwerkbreed verkeersmanagement

Intelligent Transportation Systems are a promising set of technological advancements, whose main objective is that of enabling road users and policymakers alike to better exploit the available transportation infrastructure at its best possible level of service. The reduced costs of collecting, storing and processing information are enabling the development of richer mathematical models, which in turn can be exploited to better understand and anticipate the behaviour of transportation networks. From the perspective of Dynamic Traffic Management policymakers, this is highly desirable: being able to correctly anticipate the behaviour of users travelling in a network has been historically recognized as a strict prerequisite to achieve system optimal performances. However, a gap can be found in the state-of-the-art between what’s theoretically achievable and what’s practically feasible. Existing anticipatory traffic control policies are defined in a fully centralized framework, where a single, global entity is responsible for determining the network-wide optimal control law. This is undesirable from two perspectives: on the one hand, no centralization exists in the real world where multiple (hierarchical and/or equivalent) controllers interact; on the other hand, the centralized problem is computationally complex. In the “control track” of this work, the aim is that of determining under which conditions such a centralized problem can be decomposed into smaller, distinct sub-problems. To achieve this goal, the problem is analysed both analytically from a time static point of view and empirically from a time dynamic perspective. Once the nature of the controller-to-controller interactions is established, controller-wise decomposition policies are developed, together with the respective conditions for optimality. Decomposed anticipatory control techniques are therefore introduced in form of algorithms, tested and validated through in-silico experiments. The developed methodologies are proven to be highly competitive with respect to fully centralized anticipatory control solutions, as well as very beneficial compared to standard, non-anticipatory techniques. The accuracy of transportation models, upon which the aforementioned dynamic traffic management policies are based, depend directly on the amount and quality of information that is retrieved from the underlying network. When dealing with anticipatory traffic control policies, this becomes a very stringent assumption, as information on the whole transportation network is needed to achieve system optimal behaviour. Installing and maintaining a set of sensors on a network to achieve network-wide information is though very expensive, therefore the problem of determining the minimum amount of sensors to be located on a network so to obtain this information has been defined and studied in literature. Full observability solutions to the Network Sensor Location Problem have been defined, however, the amount of sensors needed to obtain full information is still beyond what is economically viable. In the “sensing track” of this work, the main focus is thus on the problem of characterizing partial observability solutions, that is, solutions in which only a subset of all sensors needed to retrieve full information from the underlying network is available. A partial observability metric capable of capturing the indirect and partially known relationships between the observed and unobserved variables characterizing the network is developed, and exploited in simple heuristic algorithms to obtain the set and locations of sensors that maximizes the amount of partially recovered information. The implications of locating sensors according to this metric are studied in Origin/Destination matrix estimation procedures. Furthermore, the relationship between the route set enumeration technique utilized to obtain full observability solutions and the resulting quality and amount of information embedded therein is assessed empirically. The overall results show that selecting sensor locations according to the developed metric is beneficial both from the theoretical aspects of observability problems and from the practical point of view of flow-estimation methodologies.status: publishe