Solving Signal Coordination Problems Using Master- Slave Genetic Algorithms

This paper presents the design of master-slave genetic algorithms (GA) in solving signal coordination problems. When a serial GA is applied, its performance in terms of computation time diminishes as more accurate results (smaller time slices to evaluate flows and queues) of network performances are needed, or the size of signal networks increases. Because GA works with a population of independent solutions, it is easy to distribute the computational load, i.e., calculating fitness values of candidate solutions, among several processors and, thus, considerably speed-up the computation time. With a master-slave GA, a single processor (master) performs all genetic operations while a number of processors (slaves) are assigned to evaluate a set of fitness functions. The fundamental step in designing a master-slave GA is to determine the optimal number of processors. In this paper, the analytical formulation of defining the optimal number of processors and the empirical results from the master-slave GA application are presented. A master-slave GA is implemented to a signal coordination problem for a network with oversaturated intersections. For a given network size, the performance of a master-slave GA is investigated when network performances (flows and queues) are evaluated at different sample times, ∆T. When the fitness evaluation time is large (using smaller ∆T) relative to the communication time, the master-slave GA is more efficient and provides larger speed-up. The performance of a master-slave GA is also investigated when the size of signalized networks or the size of problems is changed. Results indicate that an efficient master-slave GA (larger speed-up) is attained when it is used to solve a problem with higher fitness evaluation time and to solve a network with a larger size. Although a master-slave GA only provides a lower bound speed-up, the illustration in this paper demonstrates the potential merit of using parallel GA in solving a signal coordination problem.