Application of genetic algorithms to generate optimum signal coordination for congested networks

This paper presents two different Genetic Algorithms (GA) applied to design signal coordination for oversaturated networks. Signal coordination is formulated as a dynamic optimization problem and is solved using GA for the entire duration of congestion. This paper considers a tradeoff between simple GA (SGA), which requires a large population converging in a single convergence epoch, and micro GA (MGA), which requires smaller population with multiple epochs. A comparison is made for given resources available, that is, a fixed number of function evaluations. To provide quality solutions, SGA requires a large population but takes a longer time to converge, and thus it is not efficient for a real-time system. MGA overcomes the drawback encountered by SGA, that is, the time penalty involved in evaluating the fitness values for a large population. This paper reveals that MGA implementation on signal coordination problems reaches the near-optimal region of signal timing much earlier than SGA implementation. For a given number of functional evaluations, a small population size of MGA outperforms SGA executed with a larger population size.