抄録
Recently, several heuristic computational methods have been developed for complex real problems. Especially, PMBGAs using statistical information of the good search points have been attracted attention by many researchers and several types of PMBGAs are proposed. In this paper, a novel real-coded probabilistic model-building Genetic Algorithms (PMBGAs) which is called Multiple Neighborhoods Sampling (MNS) is proposed. PMBGAs have a issue that search points converge, while the MNS has a characteristic that search points are maintained different diversities. In the MNS, a new searching point is generated using the neighborhood area. When the neighborhood is narrow, a new search point is generated for the local search. On the other hand, when the neighborhood is wide, a new search point is generated for global search. The MNS has multiple neighborhoods whose width is different respectively. Therefore, search points do not converge and the MNS can run efficiently solution search. Through the standard test functions, the effectiveness of the MNS is examined. The results describe that the MNS shows the better performance than Simple GA.
