Evolving Marine Predators Algorithm by dynamic foraging strategy for real-world engineering optimization problems

Abstract

The Marine Predators Algorithm (MPA) is a novel hunting-based optimizer. The MPA’s central concept is based on the well-known Lévy Flight (LF) and Brownian Motion (BM) strategies as well as a simple transition model between these two strategies. The canonical MPA proposes three static steps to tune the transition behavior between the LF and BM strategies. Although MPA provides exemplary performance in many test functions, the discrete transition between the two mentioned phases causes it to get stuck in local optima when faced with real-world optimization problems. In order to address this shortcoming, this paper proposes a soft dynamic transition between LF and BM to model this encounter naturally, considering the continuous nature of the transition between LF and BM in marine predators’ real life. In order to evaluate the performance of the developed Dynamic Foraging Strategy MPA (DFSMPA), twenty-nine optimization test functions, thirty complex CEC-BC-2017 functions, ten benchmarks of CEC06-2019 test suit, and ten real applicable engineering problems, including power system design, synthesis and process design, industrial chemical producer, power-electronic design, mechanical design, and animal feed ratio, are employed. The DFSMPA is evaluated against four groups of standard optimization approaches, including (1) Arithmetic Optimization Algorithm (AOA), Slime Mould Algorithm (SMA), Equilibrium Optimizer (EO), Niching Chimp Optimization Algorithm (ChOA), Henry Gas Solubility Optimization (HGSO) as recent optimization algorithms, (2) Lévy Flight GWO (LGWO) and Evolutionary Algorithms with Adaptive Lévy Mutations (EALM) as the two best dynamic Lévy-based optimization algorithms, (3) SHADE, CMA-ES, and LSHADESPACMA as the three state-of-the-art optimization algorithms, and jDE100, DISHchain1e+12, CIPDE, and EBOwithCMAR as best performing algorithms in IEEE CEC06-2019 competition. Three non-parametric statistical tests, including the Wilcoxon rank-sum, Bonferroni–Dunn and Holm, and Friedman average rank tests, are utilized to perform a comprehensive assessment. The results show that the DFSMPA achieved the first rank among 46 out of 70 benchmark functions and engineering problems and exhibited similar results compared with SHADE and CMA-ES in other benchmarks. The statistical analysis demonstrated that DFSMPA is a significantly superior optimizer than the three first categories’ benchmark algorithms, while its result is statistically similar to jDE100, DISHchain1e+12.