Some Variants of the Optimal Strategy Transit Assignment Method



Some Variants of the Optimal Strategy Transit Assignment Method

Authors

I Constantin, M Florian, INRO Consultants Inc, CA

Description

The purpose of this paper is to present two variants of the optimal strategy model which render the sensitivity of the resulting transit flows to model parameter changes smoother without sacrificing its computational efficiency.

Abstract

The optimal strategy public transport (transit) assignment method due to Spiess
and Florian (1989) has seen extended application in a wide variety of applications.
It has been extended to congested networks where a delay/discomfort functions is associated with the segments of transit lines (such as in the RAILPLAN model used
by Transport for London) and to handle both segment delay functions and vehicle
capacities (Cepeda, Cominetti and Florian (2006). It has been employed as a sub
problem in transit network design models without and with the consideration of
congestion (Constantin and Florian (1995) and Noriega and Florian (2003) respectively).
It was also studied by other researchers and compared with other simpler methods
(see for instance DeCea, Bunster, Zubieta and Florian (1988)).

Since the resulting flows are the solution of the linear program the solutions exhibit extremal properties when the travel costs are constant. While this is of little
consequence in transit networks that are subject to congestion when the line
segment travel times are constant the changes of the resulting optimal strategies
are not smooth when model parameters change. This was remarked by several
researchers and professionals in this field (see for instance the paper by Nokel and
Weckek (2007)).

The purpose of this paper is to present two variants of the optimal strategy model which render the sensitivity of the resulting transit flows to model parameter changes smoother without sacrificing its computational efficiency. The computations are still done once for each destination.

One variant is still a deterministic model however the resulting optimal strategies discriminate better between fast and slow transit services. Another variant is a stochastic assignment which provides a logit choice among strategies without needing an explicit enumeration of the simple strategies generated. This approach has much in common with Dial?s (1971) stochastic assignment method and is inspired by the work of Nguyen, Pallottino and Gendreau (1998).

Computational results with these new variants of the optimal strategy based transit assignment are presented on large scale networks.

Publisher

Association for European Transport