A Comparison between SGA and BOA Applications to Design Signal Coordination

This paper presents the comparison of Genetic Algorithms (GA) application to signal coordination for congested networks. Signal coordination is formulated as a dynamic optimization problem in which green times are decision variables and are represented in the individual GA candidate solutions. Two different types of serial GAs will be used depending on the way a set of candidate solutions is derived. The first is the Simple Genetic Algorithm (SGA). This GA uses three standard genetic operators: selection, crossover, and mutation, to generate a new set of solutions from the previous one. The second is the Bayesian Optimization Algorithm (BOA), which generates new candidate solutions using an estimate of the joint distribution of current promising solutions. The joint distribution is constructed using Bayesian networks. Investigation of each type of GA was made in terms of the number of functional evaluations needed to achieve predefined convergent criteria when the size of GA's candidate solution (population) is increased. Solution qualities are also compared. The growth of functional evaluations for SGA is close to a quadratic function with respect to the size of population, while that for BOA is a cubical function. For a small population size, SGA provides better quality solutions, while for a larger population size, BOA yields better results.