A Comparison of Operator Selection Strategies in Evolutionary Optimization

Evolutionary algorithms (EAs) are an effective paradigm for solving many types of optimization problems. They are flexible and can be adapted to new problem classes with little effort. EAs apply operators on the elements of a population. When multiple operators are involved, their distribution is based on fixed probabilities. EAs therefore can not react on changes during an optimization which often leads to premature convergence. In this paper, we present a variation of our approach described in (C. Breitschopf et. al, 2005) for a self-adapting operator selection that is able to monitor the success of the operators over time and gives priority to currently successful operators. We compare the results with another approach we implemented as first strategy for considering operator success as well as analyze under which circumstances which approach should be preferred