A distance alternation model on real-coded genetic algorithms

We propose a distance dependent alternation (DDA) model as a generation alternation model on real-coded genetic algorithms (CA) to improve its performance by maintaining diversity of populations. The basic concept of the DDA is that the elite of offspring will alter the nearest parent individual in multi parental GA. In other words, the DDA is an alternation scheme utilizing distance information between individuals. Further, we extend this concept to multiple individual alternation, called distance dependent multiple alternation (DDMA). Classifying offspring according to nearby parent, and generation alternation is taken place in each group. Thus multiple alternation will occur at the same time in each cluster. It improves GA performance in its computation times. We compared the performance of the proposed alternation model with the minimal generation gap (MGG) model proposed by (Satoh et al., 1996) in several functional optimization problems with the multi-parental unimodal normal distribution crossover (UNDX-m) that shows good performance as a crossover operator for the real-coded GA. The results show the effectiveness of the proposed alternation model in maintaining diversity of populations robustly and in improving performance on the real-coded GA