Frequency-based transit assignment aims to model the choices of transit passengers who select their itinerary based on their knowledge of routes, travel times and frequencies (or headways) of the lines in the network. Timetable information is not available. Next to the regularity of service, itis the level of assumed passenger information in which existing approaches mainly differ. While the classical Optimal Strategies algorithm is basedon the assumption that passengers do not have any departure time information, the introduction of passenger information systems motivated various models assuming a higher level of knowledge. The different knowledge levelsassumed naturally result in different share definitions for the options contained in a choice set, but there is one thing that many approaches havein common all the same: the level of information is assumed to be constant across the choice set, the only exception usually being the choice between alighting and remaining seated. It is one objective of this paper to show that choice sets often have a hierarchical structure, and how strongly this property affects the choice option shares. It is necessary not only to consider what kind of information a passenger has in principle, but alsofor which options that information is available at the same time. If, forinstance, transferring passengers can continue from one of several stops,they usually make a hierarchical decision: first they choose a stop, thenthey choose the next attractive line from that stop. The first decision is based on expected remaining wait and travel times for the stop, and not on observations of approaching vehicles on individual lines as in the second choice step. Even if a decision situation is modelled in a hierarchicalfashion, the mathematical model of the first choice step still is a delicate (and often neglected) topic. A very prominent example is the choice between alighting and remaining seated. Given a destination zone, many models assume a pre-defined alighting stop for passengers riding on a (segment of a) line. This approach has two obvious shortcomings: it cannot capture the fuzziness in passengers behaviour, and the resulting discontinuities may adversely affect demand model calibration. It is shown that such a zero-one-approach rarely yields satisfactory results. Different non-trivial solutions for the first choice step will be presented, evaluated and relatedto transit network conditions. It comes as no surprise that there is no single best solution. This makes it very important to better understand therelation between real-world network properties and modelling assumptions.Moreover, which combinations of share definitions on the one hand and choice set structuring on the other hand constitute suitable overall models for applications to real transit networks are assessed. Another objective of this paper is to analyze the effect of choice set structuring on the assignments performance in terms of computation time and result quality. For the covering abstract see ITRD E145999
Abstract