An adaptive framework to select the coordinate systems for evolutionary algorithms

Many evolutionary algorithms usually utilize the fixed original coordinate system to search and cannot effectively match different function landscapes. To solve this issue, this paper proposes an adaptive framework, named STCS, to select the coordinate systems for evolutionary algorithms. In STCS, the eigen coordinate system is constructed by an archive-based covariance matrix, which can capture the feature of the function landscape. What is more, the selection process of the original coordinate system and the eigen coordinate system is defined as a Markov decision process and is controlled by reinforcement learning algorithm. STCS is applied to three popular evolutionary algorithms, i.e., differential evolution, particle swarm optimization, and teaching-learning-based optimization. The experimental results on IEEE CEC2013, IEEE CEC2014, and IEEE CEC2017 test suites demonstrate that STCS is efficient and competitive. (C) 2022 Elsevier B.V. All rights reserved.

Automated summary

This paper proposes an adaptive framework, STCS, for selecting coordinate systems in evolutionary algorithms. By constructing eigen coordinate systems using an archive-based covariance matrix, the issue of ineffective matching of different function landscapes is addressed. Experimental results demonstrate the efficiency and competitiveness of STCS across multiple test suites.