Multi-modal transport relates to trips for which travellers use two or more transport modes, for example bicycle and train, train and bus, or private car and metro. Given the growing importance of chain mobility, there is an increasing number of planning problems that require a tool for analysing multi-modal travel choices, which must deal with a simultaneous choice of routes, travel modes and transfer locations. However, multiple-choice dimensions involved in a multi-modal trip are difficult to model. Choice analysis and prediction may be significantly improved by modelling multi-modal travel choices with an approach based on a clear distinction between a choice set generation step and, conditional on that set, the genuine choice modelling step. Choice set generation consists in finding all feasible routes that a traveller might consider for travelling from his origin to his destination. In a route choice context, the choice set composition is a critical aspect because very many routes may be available whereas only a limited subset of those are actually perceived while even less are actually considered by trip makers. In a multimodal transport network, route finding and generation is even more difficult because of the multiple different types of choices involved in a multi-modal trip. The specific theoretical challenge in modelling multi-modal trips is in the multi-dimensional character of these trips encompassing a multitude of choices with respect to routes, travel modes, transport service types, and interchange locations between public transport modes, access/egress locations from private to public transport modes and vice versa. This is the challenge that is addressed as subject of study in this thesis. The main theme in this dissertation is thus to establish a choice set generation model and algorithm, and demonstrate its validity and feasibility for demand prediction purposes. Choice set generation in the case of multi-modal trips exhibiting multiple-choice dimensions (such as transport modes, transfer nodes, public transport service types, and routes) is represented jointly as a case of route choice in a multi-modal network for predicting flows in the multi-modal network. New choice set notions and related terminologies have been developed thereby explicitly accounting for different stages in the travel choice process. The formulation of the main characteristics that adequate choice sets and appropriate generation processes should satisfy in uni-modal and multi-modal networks is the main new topic dealt with in this thesis. A large number of route generation methods proposed in literature for the uni-modal networks have been analyzed in order to indicate which of these methods are potentially suitable for generating satisfactory route choice sets. A newly developed route choice set generation algorithm has been presented. This is the so-called doubly stochastic approach applicable to uni-modal and especially multi-modal networks, in which not only the link attributes but also the preference (or behavioural) parameters of the cost function are randomized, since especially in a multi-modal context, variations in transport modes and in multi-modal combinations are desirable in the generated choice sets. This doubly stochastic method has been tested with respect to its empirical and probabilistic properties, and applied to various networks, in particular to a multimodal network. The main conclusion of this study is that the various analyses performed with the doubly stochastic choice set generation approach show that the resulting choice sets have an adequate variety of uni-modal and multi-modal route alternatives, and match the desired properties in terms of choice set size (minimum and maximum size) and composition (variety in trip properties). The face validity (plausibility of generated alternatives) and empirical validity (conformity with observations) are very high. Moreover, in a number of simulations the prediction success rates (predicting the correct sequence of legs in the chosen multi-modal trip, which means among others correct entry, exit and transfer stations, correct train type and correct sequence of public and private mode) is quite optimal. (Author/publisher)
Samenvatting