Fast perturbation-dependent reliability curves in traffic networks

Abstract

Uncertainty characterisation and reliability analysis for high-fidelity models is often prohibitive due to the large analysis efforts it demands. This is particularly prevalent in highly complex systems that require costly simulations, such as that case of traffic networks. If reliability of traffic networks is to be evaluated for different perturbations, regardless of how it is defined, then prohibitive analysis times and efforts should be expected.

Traffic networks are network systems composed of multiple sub-systems and components. When changes in the system intrinsic variables occur, these result in operational changes in the network that can only be understood in an holistic form.

In the present work, a perturbation-dependent fast reliability assessment is proposed. It considers reliability as a variation in travel time to the reference time, which is often used to characterize reliability in traffic. In the present work it is discussed in a full probabilistic context, with reliability curves being characterised using a lower-fidelity model that uses a kriging-based sequential learning approach approach. This metamodeling approach enables the characterisation of different levels of reliability for a perturbation, through a N threshold modelling approach, that uses probability density functions and that sets reliability curves in a form of a fragility curve. With such implementation it is possible to enable a fast characterisation of reliability, and its probabilistic behaviour, in traffic. The implementation is researched in two reference traffic networks with uncertain demands, and results show that this technique can inform multiple purposes of decision-making, ranging from reduced order modelling tools to operational management of the system.