Methodologies for Origin-Destination travel demand estimation within a strategic traffic assignment model

Abstract

Day-to-day demand volatility is an inherent property of transport systems, which can substantially impact network performance and must therefore be incorporated into the transport planning process. The thesis proposes methodologies to explicitly account for such demand volatility in two aspects of the transportation planning process: the traffic assignment and the estimation of Origin-Destination demand. The traffic assignment modelling contributions are based on a novel traffic assignment model - strategic user equilibrium (STRUE). Under STRUE travellers choose routes to minimize their expected travel cost, where their decision is based on knowledge of a demand distribution, rather than a deterministic demand. STRUE is defined such that at equilibrium, all used paths have equal and minimal expected travel costs. Two extensions of STRUE which relax previous simplifying assumptions and enhance the model’s applicability are proposed. For each extension, a convex mathematical program is formulated, and the model’s performance is evaluated. The numerical analysis demonstrates that the STRUE can account for demand volatility while maintaining computation efficiency. In addition to traffic assignment modelling, Origin-Destination demand estimation is addressed. A novel bi-level programming framework is proposed to estimate the total demand, which calls on the aforementioned STRUE model, thus incorporating day-to-day demand volatility into the estimation process. A mathematical proof demonstrates the convexity of the proposed framework. The numerical analysis illustrates the efficiency and sensitivity of the proposed framework. An additional challenge for estimating O-D demand matrices results from the large number of O-D pair demands to be estimated, which is often much greater than the number of monitored links. The assumption of the sparse O-D matrix can be used to address the under-determination problem. Thus, sparsity regularization is combined with link flow correlation to provide additional inputs for the O-D estimation process to improve the solution quality. The model is formulated as a convex generalized least squares problem with regularization, the usefulness of sparsity assumption and link flow correlation is presented in the numerical analysis. To summarize, the thesis generalizes the novel STRUE model to improve its applicability and proposes two robust demand estimation models while explicitly accounting for both demand and link flow variation.