Scheduling of lane closures using genetic algorithms with traffic assignments and distributed simulations

In past research, several versions of hybrid genetic algorithm-simulation methodology have been proposed for scheduling of multiple lane closures that aims to minimize a network's total traffic delay. The genetic algorithm is used as a search engine for generation of lane closure schedule, while a microscopic traffic simulation model is employed to calculate the total network travel time under each lane closure scenario. A difficulty in implementing this methodology practically is the long computing time required, due to the many simulation runs needed to evaluate the average total network travel time of each feasible schedule. This paper applies the precondition technique, standard error criterion, and termination criterion to reduce the number of necessary simulation runs. As a further improvement, traffic simulations are distributed in different processors of a multiprocessor machine. To further reduce the computing time, a two-stage hybrid genetic algorithm methodology has been proposed in this paper. This two-stage methodology consists of a hybrid genetic algorithm-traffic assignment methodology as the first stage, followed by a hybrid genetic algorithm-distributed simulation methodology as the second stage. The traffic assignment model is used to replace the traffic simulation model in the estimation of total network travel time in stage 1. The applications of the improvement techniques have been demonstrated through a hypothetical problem involving 20 lane closure requests in a network consisting of 986 links, 397 nodes, and 22 origin-destination zones. Together, these improvement techniques contributed to up to 87% reduction in waiting time for a solution of the example problem.