Steady-state link travel time methods: Formulation, derivation, classification, and unification

Abstract

Travel times are one of the most important outputs of transport planning models, especially in a strategic context. It is therefore paramount that the methods that underpin the construction of travel times are well understood. A plethora of methods exists to extract and/or construct travel times given some underlying network loading procedure, also known as the traffic flow propagation model. However, the relation between these different travel time methods and the consistency between such methods has received relatively little attention in the literature. This might in part be due to the many different traffic flow propagation models in existence, ranging from vehicle based (microscopic), to flow based (macroscopic), and models that explicitly account for the time varying nature of traffic flows (dynamic) to models that do not (static). In this work, we limit ourselves to flow based, i.e. macroscopic, traffic flow models. Within this modelling paradigm we consider dynamic, semi-dynamic, and static traffic flow propagation formulations used to construct link travel times. The semi-dynamic and static approaches are considered as more aggregate versions of the dynamic formulation. Within this context we formulate a unified (link) travel time formulation that is consistent across these three modelling paradigms under the assumption of steady-state flow conditions. The dynamic link travel time formulation is based on a recent state-of-the-art continuous time macroscopic dynamic network loading model. In the dynamic model we assume steady-state conditions to remain consistent with steady-state semi-dynamic and static approaches. This allows us to derive semi-dynamic link travel time formulations from our dynamic model, while the static formulation is derived from its semi-dynamic counterpart. Throughout this work we explore link travel times from three different perspectives; an experienced perspective, which actively tracks the tail of a physical queue, and two functional perspectives, both of which do not require explicit queue tracking. Based on the existing literature and proposed formulations, a classification framework is proposed allowing one to compare existing (and future) methods in the literature in an objective fashion. We provide a number of explicit derivations of existing model formulations that can be considered special cases of our unified approach. A numerical example across the different perspectives is included and a significant number of representative existing methods in the literature has been classified based on our proposed framework for the reader's convenience.