Multi-space evolutionary search with dynamic resource allocation strategy for large-scale optimization
Qingxia Shang,Yuxiao Huang,Junwei Dong,Yaqing Hou,Yu Wang,Min Li,Liang Feng
DOI: https://doi.org/10.1007/s00521-021-06844-4
2022-01-11
Neural Computing and Applications
Abstract:Multi-space evolutionary search (MSESdocumentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$${
m MSES}$$end{document}) is a recently proposed paradigm to optimize across multiple solution spaces for solving a large-scale optimization problem. MSESdocumentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$${
m MSES}$$end{document} allocates the computational resource equally on each search space. However, different search spaces are likely to make different contributions to the finding of global optimum of the given problem. Dividing the limited resources equally on each search space in MSESdocumentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$${
m MSES}$$end{document} is thus not an efficient strategy. This paper aims to investigate how to utilize the imbalanced efforts to allocate computational resources in multiple search spaces to efficiently improve the performance of MSESdocumentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$${
m MSES}$$end{document}. In this paper, we propose a novel multi-space evolutionary search with dynamic resource allocation strategy (MSES-DRAdocumentclass[12pt]{minimal}usepackage{amsmath}usepackage{wasysym}usepackage{amsfonts}usepackage{amssymb}usepackage{amsbsy}usepackage{mathrsfs}usepackage{upgreek}setlength{oddsidemargin}{-69pt}egin{document}$${
m MSES-DRA} $$end{document}) for large-scale optimization. In particular, a detection mechanism is presented to measure the reasonableness of assignment in terms of computation resource of different spaces. Further, according to the interaction between optimal individual and population, the proposed dynamic resource allocation strategy is designed based on the explicit–implicit contributions of spaces. The explicit and implicit contributions are defined by the fitness improvement of best solution and the survival of individuals, respectively. An adaptive technology based on the feedback is conducted to balance the assignment of computational resources for each search space. To evaluate the performance of the proposed method, comprehensive empirical experiments have been conducted on the CEC2013 large-scale benchmark problems.
computer science, artificial intelligence