Decomposition of large-scale constrained problems using a genetic-based search

Abstract

In this paper, a traditional genetic algorithm with an integer representation is developed to search suitable variable arrangements for decomposition problems into a Cooperative-Coevolution framework for large-scale constrained optimization problems. The function evaluation is based on the Variable Interaction Identification for Constrained Optimization Problems (VIIC), that detects interactions among variables in large-scale problems. The new algorithm is capable of getting variable arrangements where the number of subgroups, as well as variables in each subgroup, are not fixed as in traditional decomposition methods. The proposed method is compared against VIIC with neighborhood strategy (VIICN). Those algorithms are tested on a novel benchmark for large scale constrained problems with 100, 500 and 1000 dimensions. The numerical results indicate that the proposed method outperforms VIICN, with the computational time is greatly reduced, which makes DVIIC a viable method for solving large-scale constrained optimization problems.