Backtracking search algorithm driven by generalized mean position for numerical and industrial engineering problems

Backtracking search algorithm (BSA) is a very popular and efficient population-based optimization technique. BSA has a very simple structure and good global search ability. However, BSA may be trapped into the local optimum in solving challenging multimodal optimization problems due to the single learning strategy. To enhance the global search ability of BSA, this paper proposes an improved version of BSA called backtracking search algorithm driven by generalized mean position (GMPBSA). In GMPBSA, two types of generalized mean positions are defined based on the built feature zones, which are employed to design the comprehensive learning mechanism consisting of three candidate learning strategies. Note that, this learning mechanism doesn't introduce new control parameters and refer to the complex calculation. To verify the performance of GMPBSA, GMPBSA is used to solve the well-known CEC 2013 and CEC 2017 test suites, and three complex engineering optimization problems. Experimental results support the great potential of GMPBSA applied to the challenging multimodal optimization problems. The source code of GMPBSA can be found from

Automated summary

This paper proposes an improved version of backtracking search algorithm called GMPBSA, which introduces generalized mean positions and a comprehensive learning mechanism to enhance the global search ability of BSA. Experimental results show the great potential of GMPBSA in solving challenging multimodal optimization problems.