Continuous non-revisiting genetic algorithm with overlapped search sub-region
Chi Kin Chow, Shiu Yin Yuen · 2012
In continuous non-revisiting genetic algorithm (cNrGA), search space is partitioned into sub-regions according to the distribution of evaluated solutions. The partitioned subregion serves as mutation range such that the corresponding mutation is adaptive and parameter-less. As pointed out by Chow and Yuen, the boundary condition of the mutation in cNrGA is too restricted that the exploitative power of cNrGA is reduced. In this paper, we tackle this structural problem of cNrGA by a new formulation of mutation range. When sub-region is formulated as which certain overlap exists between adjacent sub-regions, this creates a soft boundary and it allows individual move from a sub-region to another with better fitness. This modified cNrGA is named cNrGA with overlapped search sub-region (cNrGA/OL/OGF). By comparing with another work on this problem, Continuous non-revisiting genetic algorithm with randomly re-partitioned BSP tree (cNrGA/RP/OGF), it has an advantage on processing speed. The proposed algorithm is examined on 34 benchmark functions at dimensions ranging from 2 to 40. The results show that the proposed algorithm is superior to the original cNrGA, cNrGA/RP/OGF and covariance matrix adaptation evolutionary strategy (CMA-ES).