C. Breitschopf, G. Blaschek, T. Scheidl: Operator Selection in Evolutionary Optimization, WSEAS Transactions on Information Science and Applications, ISBN: 1790-0832, Issue 1, Volume 3, January 2006.
Evolutionary algorithms are a commonly known technique for solving different types of combinatorial optimization problems. They typically work on a population of solution candidates and create new solutions based on existing ones by using appropriate operators. Solutions that should be “modified” are selected based on their fitness. The selection of operators requires smart strategies as it strongly influences the quality of the final solution and the time for finding good solutions. In this paper, we present a novel approach for operator selection and evaluation during an optimization run. The basic idea behind our strategy is to estimate the future success based on the experience from the past. This concept leads to a permanent adaptation to the operator success over time so that an appropriate operator can be chosen for a given solution at every point in time. We integrated our strategy in the OptLets framework to demonstrate the flexibility and results we achieved for different optimization problems. As our approach considers evolutionary concepts in general, it can also be used for other population-based optimization systems.