A problem decomposition approach for large-scale global optimization problems

A V Vakhnin,A S Polyakova,D V Kustov,E A Sopov,I A Panfilov

doi:10.1088/1757-899x/537/5/052031

Abstract

In fact, many modern real-world optimization problems have the great number of variables (more than 1000), which values should be optimized. These problems have been titled as large-scale global optimization (LSGO) problems. Typical LSGO problems can be formulated as the global optimization of a continuous objective function presented by a computational model of «Black-Box» (BB) type. For the BB optimization problem one can request only input and output values. LSGO problems are the challenge for the majority of evolutionary and metaheuristic algorithms. In this paper, we have described details on a new DECC-RAG algorithm based on a random adaptive grouping (RAG) algorithm for the cooperative coevolution framework and the well-known SaNSDE algorithm. We have tuned the number of subcomponents for RAG algorithm and have demonstrated that the proposed DECC-RAG algorithm outperforms some state-of-the-art algorithms with benchmark problems taken from the IEEE CEC’2010 and CEC’2013 competitions on LSGO.

Highlights

Today, there are a lot of relevant real-world optimization problems that involve many variables into optimization, for example [1]–[6]
We have described details on a new DECC-RAG algorithm based on a random adaptive grouping (RAG) algorithm for the cooperative coevolution framework and the well-known SaNSDE algorithm
We have tuned the number of subcomponents for RAG algorithm and have demonstrated that the proposed DECC-RAG algorithm outperforms some state-of-the-art algorithms with benchmark problems taken from the IEEE CEC’2010 and CEC’2013 competitions on large-scale global optimization (LSGO)

Summary

Introduction

There are a lot of relevant real-world optimization problems that involve many variables into optimization, for example [1]–[6]. We have tuned the number of subcomponents for RAG algorithm and have demonstrated that the proposed DECC-RAG algorithm outperforms some state-of-the-art algorithms with benchmark problems taken from the IEEE CEC’2010 and CEC’2013 competitions on LSGO.

Results

Conclusion