Adaptive Dimensionality Reduction in Multiobjective Optimization with Multiextremal Criteria

摘要

The paper is devoted to consideration of multicriterial optimization (MCO) problems subject to multiextremality of criteria. Application of convolution techniques for finding partial Pareto-optimal solutions generates under this assumption the multiextremal problems of scalar optimization. For solving these problems it is necessary to use efficient global optimization algorithms. As such the methods the nested schemes of dimensionality reduction in combination with univariate characteristical optimization algorithms are considered. A general description of the scheme is given and its modification accelerating the search is presented. Efficiency of the proposed approach is demonstrated on the base of representative computational experiment on a test class of bi-criterial MCO problems with essentially multiextremal criteria.