Backtracking search algorithm with specular reflection learning for global optimization

Benefiting from population, randomness and simple structures, metaheuristic methods show excellent performance for solving global optimization problems. However, in some cases, in order to get promising solutions, the existing metaheuristic methods usually need to be modified. This work reports a new technique, called specular reflection learning, for improving the optimization performance of metaheuristic methods. Specular reflection learning is motivated by specular reflection phenomenon in physics. Note that, there is a close relationship between opposition-based learning and specular reflection learning. Opposition-based learning can be seen as a special case of specular reflection learning. In order to investigate the effectiveness of specular reflection learning, specular reflection learning is employed to improve backtracking search algorithm (BSA). The performance of the proposed backtracking search algorithm with specular reflection learning is evaluated by 88 test functions extracted from the well-known CEC 2013, CEC 2014 and CEC 2017 test suites, and two constrained engineering design problems. Experimental results confirm that specular reflection learning is a more effective technique for improving BSA compared with opposition-based learning, which establishes the foundation for the applications of specular reflection learning on other metaheuristics. In addition, the source code of this work can be found from https://www.mathworks.com/matlabcentral/fileexchange/79030-bsa_srl. (C) 2020 Elsevier B.V. All rights reserved.

Automated summary

The study introduces a new technique called specular reflection learning to improve the optimization performance of metaheuristic methods, particularly in enhancing the backtracking search algorithm. The effectiveness of specular reflection learning is demonstrated through experiments with various test functions and engineering design problems, showing its superiority over opposition-based learning.