< Back to previous page

Project

Intersection Modelling and Marginal Simulation in Macroscopic Dynamic Network Loading

The ever increasing demand for mobility causes traffic jams and delays on road networks all over the world, generating immense social and economic losses. Dynamic Network Loading (DNL) models represent the propagation of traffic and congestion. This PhD thesis focuses on macroscopic simulation-based DNL modelling based on first-order traffic flow theory (Lighthill & Whitman, 1955; Richards, 1956). These models support the decision-making of road managers and the information provision to road users.</>
 </>
The general aim of this PhD research is to further advance first-order macroscopic DNL models and their practical applicability. More specifically, this thesis pursues two quite distinct researchdirections:</>
 </>
1.      </></></>Enhancing the theoretical knowledge and soundness as well as the realism and practical applicability of macroscopic DNL intersection models</>
2.      </></></>Developing marginal DNL simulation models, which combine significant computation time savings with</>realistic congestion dynamics</> in repeated (iterative, finite difference or Monte-Carlo) DNL simulations</>
 </>
The function of the intersection model in the DNL model is twofold. Firstly, it must find a consistent solution for the traffic flows (veh/h) from each incoming to each outgoing link, considering the external constraints from these links (i.e. the local demands and supplies). The second function is toimpose internal supply constraints, i.e. additional flow restrictions due to conflicts internal to the intersection (e.g. crossing streams hindering each other).</>
From a thorough literature review, it is concluded that the vast majority of existing models fails to properly fulfil the first function. In response, we compile a list of seven generic requirements for first-order macroscopic DNL intersection models and present a general intersection model (limited to external constraints) that meets all of these requirements.</></>
Since most existing models neglectinternal supply constraints, they are not well suited for busy urban and regional intersections. In this thesis, it is proposed to introduce internal supply constraints analogous to how external supply constraints are universally treated in the state-of-the-art. This implies that the internal supplies are distributed according to the proportionality of predefined priority parameters of the incoming links. The possibility of non-unique solutions is identified (in general; not only in our model specification). It is found that solution uniqueness is only guaranteed if the priority parameters are single-valued. This implies that all movement flows from an incoming link have the same competitive strength for all the internal and external conflicts of the intersection. Since this uniqueness condition is intuitively contradictory to the observation that priorities often differ per conflict (for instance a straight movement usually has priority, while a left turn has not), it hinders the definition of the intersection model.</>
Specific intersection models for different types of intersections are developed that solve this ambiguity via aweighted pre-processing of different priority parameters per conflict into a single representative value. While there are still several ways tofurther improve these models, they are to the best of our knowledge -</>the first to combine both functions of the intersection model into aunique, consistent solution.</></>
 </>
Secondly, the novel concept of marginal DNL simulation is introduced. </>Marginal DNL algorithms are derived from a maternal base model, from which they adopt the modelling principles and (the majority of) the simulation algorithm. Theyperform partial (marginal) simulations of local variations to a base scenario, rather than running full simulations. This provides a considerable computational advantage if many repeated simulations with large overlap need to be performed. </></>
In this thesis, two marginal DNL algorithms are proposed. For both, the maternal base model is the Link Transmission Model (LTM) of Yperman (2007). Hence, the congestion dynamics inboth algorithms are consistent with first-order traffic flow theory as in Newell (1993). </>
The first, the Marginal Incident Computation (MIC) algorithm, is designed for fast Monte-Carlo simulation of incidents.Compared to LTM, MIC may reduce the computation time to less than 1 % (depending on the network size), at the cost of acceptable approximation errors (of aggregated outputs such as vehicle hours lost). </>
Secondly, the Marginal Computation (MaC) algorithm is presented. </>MaC has anextended functionality (both demand and supply variations) and higher accuracy compared to MIC, enabling analysis of fine-grained output such as (link) flows. </>Possible applications include </>variability studies,optimization</> problems such as dynamic origin-destination estimation and optimal control, robust network design and (real-time) dynamic traffic management support; all of which are </>currently infeasible or at least highly computationally demanding.</>
</>Finally, the applicability of marginal simulation can be enhanced by tailoring specific marginal algorithms </>for specific purposes. Moreover, we estimate that the advantage of marginal simulation could carry over to other research domains with similar needs (for instance pedestrian modelling and supply chain management). In fact, it has turned out that very similar techniques havebeen used for the </>design of digital hardware circuits (Hwang et al.,1988; Salz & Horowitz, 1989).</>
Date:3 Mar 2008 →  23 May 2012
Keywords:Traffic flow model
Disciplines:Urban and regional design, development and planning
Project type:PhD project