Adaptive Genetic Algorithms: A Methodology for Dynamic Autoconfiguration of Genetic Search Algorithms

Genetic Algorithms (GA) like other modern metaheuristics claim to be general problem solvers. Though GA have been applied successfully to a wide range of different combinatorial optimization problems, the need for careful and time-consuming tuning of GA constitutes a major drawback as a general problem solver. In this report we introduce the concept of Adaptive Genetic Algorithms (AGA) as a solution to this calibration problem, which dynamically performs an on-line autoconfiguration of the GA-parameters. To demonstrate the superior performance of AGA vs. GA in terms of solution quality, robustness and computational effort, we present computational results for three different combinatorial optimization problems. Our benchmark comprises two standard benchmark problems (Quadratic Assignment Problem and Period Vehicle Routing Problem) and one real-world problem arising in airline scheduling.