摘要

Conditionally breeding real-coded genetic algorithm (CGAR) is effective approach for continue domain problems, in which crossover and mutation behaviors are performed by difference-degree between individuals instead of given probability. In this paper we present a novel exploitation scheme for CGAR to balance between two contradictory aspects of its performance: exploration and exploitation, which is aimed at improving its ability to converge to the near-optimal solutions. The proposed algorithms are evaluated on a number of benchmark functions and the simulation results show that the proposed algorithm performs quite well and outperforms other algorithms.