Traffic Assignment for Urban Evacuation

Abstract

Urban emergency evacuations impose sudden and extreme demands on road networks that differ fundamentally from normal operating conditions. When a large population must leave a defined area within a short time window, traffic demand surges abruptly, evacuation corridors are stressed toward their capacity, and the consequences of poor traffic management extend beyond inconvenience to a direct threat to public safety. Conventional traffic planning tools, which assume steady state demand, time invariant network conditions, and independently optimised intersections, are inadequate in this context. This thesis develops and evaluates a two-stage procedure for traffic assignment and signal control under emergency evacuation conditions, in which a static assignment calibrates the signal plans and a microsimulation evaluates them. Macroscopic traffic flow theory, static and dynamic assignment methodologies, and signal timing design are systematically addressed, culminating in a microsimulation case study of the Sandy Hill neighbourhood in Ottawa, modelled in Aimsun Next. Evacuation demand is derived from 2021 Census data, with a documented extrapolation from the 2011 Census counts, through a chain of population, household, vehicle ownership, and compliance parameters, yielding 3,530 evacuating vehicles loaded over a two hour horizon by a Rayleigh departure time distribution. The network comprises 17 signalized intersections organised around a two branch evacuation corridor serving the two principal egress destinations, Highway 417 and the Macdonald–Cartier Bridge. Two fixed time signal control strategies are designed from the assigned peak interval flows and evaluated: a Webster baseline plan with pedestrian governed minimum greens, and a Coordinated Green Wave plan with a common corridor cycle and progressive offsets calibrated to the loaded corridor platoon speed of 25 km/h. The comparison shows that the value of coordination depends on how heavily the corridor is loaded. While the network is lightly loaded the two plans are indistinguishable; through the peak of the Rayleigh loading profile the coordinated plan reduces average travel time per vehicle by 21.4 %, and as demand falls the two converge again. Over the whole run, across ten replications per plan with common random seeds, coordination reduces total system travel time by 9.5 %, stopped time by 15.8 % and mean queue by 33.5 %, every measure significant at the 0.1 % level. The experiment is then repeated at three higher demand levels to test whether the benefit grows with load. It does, up to a point. At 1.25 times the base demand the reduction in total travel time is 34.9 %, at 1.5 times 63.9 %, and at twice the demand it falls back to 37.9 %. The reason is visible in the two plans separately: under coordination the travel time per vehicle is 96, 95 and 103 s across the first three levels, essentially unchanged while the demand rises by half, whereas under isolated timing it is 107, 145 and 285 s. At twice the demand both plans are beyond the corridor's capacity and the gap narrows because both are failing. At 1.5 times the difference is qualitative rather than quantitative: every coordinated run cleared the network within the horizon and no isolated run did. Coordination also makes the outcome between six and twenty times more repeatable, and the margin widens with load. The saving arises not from added capacity but from the alignment of green times between junctions, and it is concentrated on the southbound route to Highway 417, where isolated timing is liable to break down altogether. The results indicate that an evacuation signal plan should be judged by its performance at the crest of the demand surge, by how reliably it performs there, and by whether the network clears, rather than by its average delay over the evacuation period, which a steady-state assessment cannot provide.