Self-Adaptive Inversion Mutation - Self-Adaptive Heuristics for Evolutionary Computation - page 89

5. Parameter k is usually decreasing during the optimization as less and less
edge swaps are advantageous for the improvement and not the destruction
of solutions. When reaching k 1, SA-INV slows down as solutions with

k> 1 are possible, but deteriorate the optimization. We call this problem
strategy bound problem . Like in the case of step sizes for ES too big mutations
destroy solutions and hinder the evolutionary process from approximating
the optimum.
6. To overcome the strategy bound problem in later search phases, we introduced
the SA-INV-c, a heuristic that sets k = 1 constantly if the condition u ≥

≈

λ

2 is
fulfilled. The modification improves the performance of SA-INV as the small
mutation strength k = 1 is essential for successful solutions in the course of
the optimization process.