Computational issues in micro-genetic algorithms for traffic management

Emerging artificial intelligence techniques such as genetic algorithms (GAs) allow a more realistic representation and solution of difficult and combinatorial problems such as the dynamic traffic queue management problem. Computational experience in solving such complex large-scale problems by use of micro-GAs is described. In addition to providing evidence of the ability of micro-GAs to successfully identify optimal traffic management schemes, some micro-GA-associated computational issues that warrant attention are highlighted. Choosing a proper population size is a critical decision, and internal variability must be accounted for to assess the goodness of the optimization results properly. A simple rule for deciding the best population size for micro-GAs is proposed. Micro-GAs may converge to low-quality solutions, particularly with very small population sizes; convergence of micro-GAs by itself is not a sufficient indication of good performance. The size of the search space for some real-world systems can pose some difficulties to micro-GAs. Choosing between micro-GAs and regular GAs is a problem-dependent decision.