Modelling and Evaluation of Reliability Impacts in Road Networks: Concepts and Methods for Traffic Assignment Models
D P Watling, ITS, University of Leeds, UK
Based on completed research studies, a state-of-the-art review is given of methods for modelling reliability in road networks, with illustrative examples and advice for their practical application.
It has long been appreciated that transport systems such as road networks do not always operate in the same state, even if we hold constant both policy measures and external influences on demand. Nevertheless, the models used in widespread practice have typically neglected such variations in the level-of-service, with the remit of transport planning to plan for the ?average? rather than the atypical. However, as growth in populations and travel demand mean that more locations in our transport networks reach capacity, the rationale for ignoring such variability is more difficult to justify, since more of the network operates in ?sensitive? regions where small changes can be magnified into large impacts on delay. In the last decade, we have thus seen a strongly growing interest in measuring, modelling and predicting road network reliability, with major initiatives underway in the UK, Netherlands, US and Japan. The purpose of the paper is to provide a state-of-the-art review of the techniques now available, to explain and illustrate their scope with simple examples, and to consider the extent to which they are currently implementable with existing network assignment packages. This is based on several recent projects that the author has been involved with, as well as a wider review of developments internationally.
In fact, the field of ?network reliability? has provided a diverse range of techniques, aimed at several different kinds of problem and application. A common theme running through such techniques has been the natural representation of (stochastic) uncertainty in the level-of-service provided by the network to the travellers using it, yet one may identify at least five classes of method that have arisen:
(a) Connectivity reliability methods, whereby each link is assigned a failure probability and the objective to compute the probability of an O?D movement being connected.
(b) Travel time reliability methods, which explore the probability distribution of O-D or network performance measures under variations in the travel times, O-D matrix and/or link capacities.
(c) Capacity reliability methods, where the aim is to determine the range or distribution of O-D demands that allow the capacity to function within its capacity, according to some given probability.
(d) Behavioural reliability methods, whereby the effect on mean network performance and/or economic appraisal measures is represented of drivers? behavioural choice response to travel time variability.
(e) Potential reliability methods, based on identifying vulnerable elements of a network under pessimistic assumptions.
Methods in classes (a)?(c), and to some extent (e), are similar in the sense that they are directly concerned with examining extreme, unusual or undesirable performance: from the point of view of a ?network manager? wishing to ensure a certain minimum level-of-service they therefore hold some considerable interest. For example, a city authority may adopt them to plan for, say, the impact of extreme weather events or just the normal effect of daily variation in demand. A similar applicability arises on an inter-urban level; and interesting such example in the UK is the Public Service Agreement Target that has been set for the Highways Agency, in maintaining a stable level of service for the worst 10% of journeys.
On the other hand, the methods in class (d) has attracted greater interest from those wishing to appraise the potential benefits of schemes that may improve reliability. In such cases, the interest is not so much in planning for unusual/extreme events ? the focus remains very much on the traditional planning remit of examining long-run average conditions. The difference comes in the fact that the travellers in the transport system (rather than the planner) are assumed to form their own views regarding unreliable or variable performance, and then this affects their mean behaviour. This pragmatic approach is convenient in that it allows us to make minimal changes to conventional economic appraisal methods; an example of such an approach is that supported by several DfT-funded studies in the UK, and the natural focus of their draft Webtag guidance.
Through simple examples, the rationale and methodology behind these classes of method is explained. Where available, practical methods are described for resolving problems concerned with estimating performance distributions over stochastic networks (relevant to (a)?(c)), or for representing traffic assignment with variable or risky alternatives in the case of (d). The paper concludes with identifying the challenges remaining to achieve widespread application of such methods, and suggests research and development priorities for the future.
Association for European Transport