evolutionary algorithm

1060 Words5 Pages
There are a few papers that use multivariate EDAs on
DOPs. In some papers like [28] are use univariate EDAs in continuous environments and [38] is another paper that uses
EDAs in discrete environments. Besides, variant Particle
Swarm Optimization (PSO) algorithms proposed on the DOPs provide good results. Therefore, to compare the results of MAMEDA we use [29] and [62]. PSO-CP algorithm [29] are utilized a new PSO model, called PSO with composite particles to address DOPs. In [62] is proposed a MA which hybridizes PSO with a fuzzy cognition local search technique on DOPs.
The experiments are divided into four groups. In the first group, we try to produce different dynamic environments to evaluate the performance of MA-MEDA. With combination of the following parameters, different conditions are produced.
Dimensions are set to 2,5,10. Number of peaks and change severity of environment are in set 1,10,100 and 1.0,2.0,5.0 respectively. The experimental results of MAMEDA are discussed in second group. Hence, the ability and weakness of algorithm are investigated. Therefore, we can evaluate the flexibility and performance rate of MA-MEDA. If the algorithm proposed is sensitive to some parameters, we discuss variant methods to improve the performance of it. We can also discuss the influence of many parameters like diversity rate, mutation rate, number of peaks and other parameters on our algorithm. So that, it is decided which the
MA-MEDA is needed to tune parameters. The results of MAMEDA are compared with PSO-CP algorithm.
The third group includes sets of experiments on the effect of correlation parameter ? on the performance of MA-MEDA. In final group, we have comparing with MA is proposed in [62].
C. Comparing MA-MEDA in Dynami...

... middle of paper ...

...e are not enough generations for algorithm to converge accurately.
Another condition influences the result of algorithm, is the number of offspring which are produced for each cluster. The produced offspring for each cluster is depend on the covariance matrix for individuals in cluster ,. If not enough offspring are produced, then these offspring do not cover the cluster , well. Therefore, the movement speed to optima will decrease. If the numbers of offspring produced increase, the evaluation numbers will rise in each generation. So, we must find the proper number of offspring for each cluster. In the experiments are carried out, the offspring number for each cluster is equal to the number of primarily individuals in that cluster. If the number of primarily individuals in the cluster is fewer than ten, the numbers of offspring produced are considered equal to ten.
Open Document