A genetic algorithm with neutral mutations for solving nonstationary function optimization problems
Kazuhiro Ohkura, K. Ueda · 2002
An extended genetic algorithm for solving nonstationary function optimization problems is presented. When using standard genetic algorithms, it is so difficult to deal with problems in which the population often fails to find or follow the changing optimum. This is due to the brittleness caused by the fact that the population tends to stay where it believes to be optimum. In order to overcome this unwanted phenomenon, a new string representation associated with inactive regions which enables one to adopt various types of neutral mutations is introduced. It is emphasized that this mechanism works effectively after obtaining directed evolution as an adaptive strategy. A 17-object knapsack problem is examined to discuss the dynamics of the extended genetic algorithm.>