Memory backtracking strategy: An evolutionary updating mechanism for meta-heuristic algorithms

Abstract

The search domain of meta-heuristic algorithms is always constantly changing, which make it difficult to adapt the diverse optimization issues. To overcome above issue, an evolutionary updating mechanism called Memory Backtracking Strategy (MBS) is proposed, which contains thinking stage, recall stage, and memory stage. Overall, the adoption of the MBS enhances the efficiency of MHSs by incorporating group memory, clue recall, and memory forgetting mechanisms. These strategies improve the algorithm's ability to explore the search space, optimize the search process, and escape local optima. MBS will be applied to three different types of MHS algorithms: evolutionary based (LSHADE_SPACMA), physical based (Stochastic Fractal Search, SFS), and biological based (Marine Predators Algorithmnm, MPA) to demonstrate the universality of MBS. In the experimental section including 57 engineering problems, algorithm complexity analysis, CEC2020 Friedman ranking, convergence curve, Wilcoxon statistical, and box plot. Among them, 21 algorithms participated in the Friedman experiment, including MBS_LSHADE_SPACMA ranked first, LSHADE_SPACMA ranked second, MBS_MPA ranked 6th, MPA ranked 8th, MBS_SFS ranked 9th and SFS ranked 12th. Combined with the analysis of "MBS testing analysis" and the experimental results of engineering problems, it has proven that MBS has universality and good ability to improve optimization algorithm performance. The source codes of the proposed MBS (MBS_MPA) can be accessed by https://github.com/luchenghao2022/Memory-Backtracking-Strategy