Challenges for Contemporary Evolutionary Algorithms
Thomas Bartz–Beielstein, Mike Preuß, Karlheinz Schmitt, Hans–Paul Schwefel, Hans–Paul Schwefel · 2010
Does one need more than one optimization method? Or, stated differently, is there an optimal optimization method? Following from the No Free Lunch theorem (NFL, Wolpert and Macready [1]), in the general case—without clearly specified task—there is not. For every single task, creating a specialized method would be advantageous. Unfortunately, this requires (i) a lot of effort, and (ii) extensive knowledge about the treated problem, and is thus not practiced. Alternatively, two strategies are usually followed when tackling a ‘new ’ optimization problem: – Adapt an existing algorithm to the problem in its current form, and/or – model/formulate the problem appropriately for an existing algorithm. The first strategy modifies the algorithm design, whereas the second strategy modifies the problem design. These designs will be discussed in detail in the remainder of this article. Whereas ‘traditional’ mathematical optimization approaches mostly favor the second approach, it may provoke unwanted side-effects: One has to make sure that the most important features of the original