| Uncertainty in transportation networks is wide spread, thus the network parameters such as demand, link capacities, and travel cost are not easy to obtain. The uncertainties in these parameters can have negative implications on infrastructure investment decisions and transportation planning and operations. For example, the uncertainty in future demand reduces the expected benefits of the network improvements. The analysis with the point estimates of the parameters can be misleading and thus can result in significant economic loss. Introducing flexibility in network modeling can alleviate the negative implications of uncertainty. Transportation network flexibility can be defined as the ease with which a network can adjust to changing circumstances and demands, both in terms of infrastructure and operation. Flexibility should be considered from the perspective of planner, users, and operation manager. Planner's flexibility is particularly important in infrastructure investment, because such investments are huge and irreversible. Advanced traveler information systems (ATIS) can increase users flexibility by providing online information to the users. In the presence of ATIS systems, the conventional traffic assignment methods---that assume users pre-select path between a desired origin and destination---are not appropriate. Additionally, network users are heterogeneous, thus the responses by all users will be different for the same information. Thus, it is necessary to develop traffic assignment methods that incorporate heterogeneous responses from users to the online information. The main focus of this dissertation is on developing frameworks for (1) flexible network capacity improvements strategy, and (2) a static traffic assignment for transportation networks with online information systems.; The study begins with providing a taxonomy for the transportation network flexibility and give definitions of the many dimensions of the flexibility. Two dimensions of flexibility are studied in more detail: (i) planners temporal flexibility, and (ii) users recourse. The former is defined as the ability to sequence investment over time, and the latter is defined as the ability to adapt to new path en-route. In the former dimensions, multi-stage stochastic network design problem, called flexible network design problem (FNDP) has been studied. For the FNDP formulations in this study, demand is assumed to be stochastic; but capacity and travel times are assumed to be deterministic. The focus of the latter dimensions is on developing stochastic user equilibrium with recourse (STOCH-UER). STOCH-UER traffic assignment incorporates non-uniform users' responses toward the online information about network condition. In this problem, the demand is assumed to be deterministic, but capacities and travel costs are assumed to be stochastic.; Three approaches are proposed to model demand stochasticity in FNDP: (a) longitudinal stochasticity, (b) latitudinal stochasticity, and (c) two-way stochasticity. All FNDP variations are formulated as stochastic mathematical program with equilibrium constraints (STOCH-MPEC). Initially, the FNDP formulations are developed for inelastic demand and then extended to elastic demand. The results of FNDP on test networks show that FNDP can give up to 80% more benefits than that of single-stage network design problem. The sensitivity analysis of FNDP reveals interesting but paradoxical behavior. It is observed that the objective function is not smooth and monotonic over a range of budget values, but sudden drops are observed at certain budget values. This is most likely because of the non-convexity of FNDP formulations. FNDP formulations are very difficult to solve, mainly because (a) the formulations are non-convex, (b) stochasticity in network parameters, and (c) network size. To efficiently solve FNDP problems, a sampling method called sample average approximation (SAA) method is used to find an approximate solution to... |
