A Building Block Favoring Reordering Method for Gene Positions in Genetic Algorithms

摘要

This work proposes an algorithm to speedup convergence of genetic algorithms that is based on the investigation of neighbouring gene values of the successful individuals of the chromosome pool. By performing some statistical inference on neighbour gene values, coherent behaving genes are detected. It is claimed that those coherent acting genes are belonging to the same building block that leads to the solution. It is desirable to keep genes of the same building block together to let them―probabilistically― live together in next generations. Therefore, a reordering of the gene positions is performed, periodically. The proposed algorithm is implemented for a minimum area non-overlapping rectangle placement problem and results are compared to a classical GA implementation.