Genetic algorithm approaches for addressing unmodeled objectives in optimization problems

Public sector decision-making typically involves complex problems that are often not completely understood. In these problems, there are invariably unmodeled issues that can greatly impact the acceptability of solutions. Modeling to Generate Alternatives (MGA) is an approach for addressing unmodeled issues in an optimization context. MGA techniques are used to generate a small number of good, yet very different, solutions to optimization problems. Because these solutions are different in decision space, they may differ considerably in performance when unmodeled objectives are considered. Many problems are sufficiently complex that traditional optimization solution procedures, and therefore traditional MGA techniques, are not readily applicable. Two techniques for performing MGA using genetic algorithms are investigated and compared. One of these techniques, which uses specialized MGA operators, is shown to produce solutions that are both better in quality and more different. This technique is also demonstrated for a realistic air quality management problem.