Aggregation in transport networks for a flexible assignment model

Steijn, Justin van (2016) Aggregation in transport networks for a flexible assignment model.

Abstract:Transport modelling is a frequently used tool for transportation planning. This research focusses on the assignment step of the four-step transportation model: we develop a flexible aggregated assignment method for transport networks in order to reduce its computational costs. Introduction In the last decades, various developments played an important role in transport modelling. First, transport policies have become more complex which has lead to more complex and time-consuming transport models. Secondly, there is an increasing demand for better consistency between different transport models describing the same geographical area. Thirdly, there is demand for flexible transport model methods that can be used to match with the detail that is required for the output. More simple and faster models are developed in order to overcome the problem of complexity and computational costs. However, these faster models cannot replace the original full-scale transport models due to lack of detail. Therefore, the goal of this research is to develop a static transport assignment model that uses smart forms of aggregation to reduce computational costs and maintain accuracy as much as possible. When aggregation is applied, potential research problems exist in the form of interpretability of results, the definition of zones and the level of aggregation. The goal that the aggregated model must achieve is two-fold: to produce accurate traffic assignment traffic volumes and travel times, for lower computational costs. To remain interpretable, the aggregated model must use well-known existing low-level zone structures and the output must be consistent with current models. Aggregation principles The shortest path algorithm used in this research is Dijkstra’s algorithm. A sub-problem is defined as one Dijkstra’s algorithm with one source and multiple targets. The size and computational cost of this subproblem depends on the number of nodes or links that is being considered before the algorithm is finished. Dijkstra’s algorithm can stop when all targets have been found. Different basic aggregation methods have been tested. Basic zone and network aggregation methods result in equally optimal computational costs and accuracy. The rest of this research focusses on zone aggregation, as it is better able to model traffic flows on all roads and using the original zone structure. Aggregation alternatives We developed the following two building blocks for aggregated assignment models: the aggregated zone hierarchy and the route reconstruction method. The zone hierarchy can either be fixed aggregated zones or adaptive zoning. With fixed zones, a single layer of aggregated zones is used in the assignment model. With adaptive zoning, there are multiple layers of aggregate zones; the idea is that every original zone (centroid) interacts with a different set of aggregated zones: small original or aggregated zones nearby and big aggregated zones at distance: this is the neighbourhood. In both zone aggregation methods, the route between any origin-destination pair is not determined directly, but must be reconstructed. The route reconstruction method can either be with first/last mile routes within aggregated zones and shared routes between aggregated zones, or two-sided route reconstruction which means that two routes are generated and reconstructed into one route. Three feasible combinations of building block options are developed: fixed zones with first/last mile routes, fixed zones with two-sided routes and adaptive zoning with two-sided routes. Conclusion and recommendation The aggregation alternatives have been applied on the transport network of The Hague. Fixed aggregated zones with first/last mile routes generate far less accurate link loads and travel times than fixed aggregated zones with two-sided routes, for only slightly better computational time. For the other two methods, the difference in computational cost and accuracy differs much between level of aggregation but less between aggregation method. Therefore, the following recommendations with regards to the application of aggregated transport models are done. When the goal is to produce accurate skim matrices, adaptive zoning with two-sided routes on a high level of aggregation is advised. The case study showed that adaptive zoning with two-sided routes on the highest aggregation level results in 97% of all origin-destination pairs with a relative travel time error of at most 2%. Adaptive zoning with two-sided routes is sometimes also the preferred way of aggregation for producing assignment results, especially when the network is big, considered in its totality and when the available time is low. For adaptive zoning on the high level of aggregation, 74% of all links have a relative link load difference of at most 10% for a reduction of 45% in computational costs. When more detail is required, for example in a variant study, fixed zoning with two-sided routes on the low level of aggregation is advised. This method results in 85% of all links within the acceptable link load difference of 10%, for a computational cost gain of almost 30%.
Item Type:Essay (Master)
Faculty:ET: Engineering Technology
Subject:56 civil engineering
Programme:Civil Engineering and Management MSc (60026)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page