Improving performance of genetic local search by changing local search space topology

The paper describes a modification of genetic local search consisting in changing local search space topology. The modification may be interpreted as an extension of the idea of distance preserving crossover used in some efficient implementations of genetic local search. In the presented approach, local optimization performed after recombination does not change elements that were common to both parents. A two phases local optimization is also considered. In the second phase, local optimization is allowed to change all elements. The proposed approach is applied to travelling salesperson problem. Results of experiments reported in the paper prove that this kind of modification of the local search space topology significantly improves performance of genetic local search.