Search

As optimization problems grow increasingly complex, characterized by their intricate difficulty, larger-scale, and diverse constraints, swarm intelligence optimization algorithms have emerged as an effective solution for addressing these multifaceted challenges. Among these, the marine predators algorithm, a recent innovation in intelligent optimization algorithms, has demonstrated remarkable efficacy in solving optimization issues. However, its application to complex CEC test function sets and engineering constraint problems reveals several limitations, including limited adaptive ability, low optimization accuracy, and high local shackle probability. This paper proposes an enhanced version of the marine predators algorithm designed to overcome its inherent shortcomings. The enhancement begins with the integration of a learning automata guided teaching–learning search mechanism during the marine memory stage. This adjustment aims to strike a better balance between exploration and exploitation across different iteration periods. Subsequently, the introduction of a logarithmic spiral exploration mechanism phase strengthens the algorithm’s ability to conduct nuanced searchers around the optimal solution, thereby improving convergence accuracy. Finally, an improved adaptive relative reflection strategy is added at the end of each iteration to enhance the algorithm’s capability to escape local optima and reduce the risk of local shackling. The optimization performance of this refined algorithm is evaluated through parameter sensitivity analysis, determining the optimal parameter values. To validate its effectiveness, the improved algorithm undergoes testing against six benchmark algorithms, including the basic marine predators algorithm and its variants, as well as other improved algorithms and those recognized with awards in the CEC2017 test suite across 100 dimensions. The evaluation focuses on optimization accuracy, the Wilcoxon rank sum test, and boxplot analysis. The test results indicate that the improved algorithm proposed in this paper outperforms the other six benchmark algorithms in optimization precision, convergence rate, and solution stability, particularly when solving complex functions in high-dimensional (100 dimensions) spaces. Furthermore, the applicability and superior performance of the improved algorithm are demonstrated through comparative analysis with four established algorithms on challenging engineering design optimization problems. These include welded beam design, process synthesis, heat exchanger network design, and design optimization of industrial refrigeration systems. The findings unequivocally showcase the enhanced algorithm’s exceptional ability to solve various engineering constraint problems effectively.