Adaptive mutation operator cycling

Parameter tuning can be a lengthy and exhaustive process. Furthermore, optimal parameter sets are usually not only problem specific, but also problem instance specific. Adaptive genetic algorithms perform parameter control during the run, thus increasing algorithm performance. These mechanisms may also enable the algorithm to escape local optima more efficiently. In this paper, we describe the fitness landscape for permutation based problems, and define local and global optima, as well as the notion of adjacency of solutions. Using these definitions we show why it makes sense to combine multiple genetic operators adaptively, give examples of this, and show that an algorithm combining multiple mutation operators has a greater chance of escaping local optima. We then describe the adaptive tournament genetic algorithm (ATGA) which uses multiple mutation operators, describing a variety of used adaptation mechanisms and conclude the paper by showing experimental results.