Modified starfish optimization algorithm based on pheromone attraction strategy and its engineering application

Download PDF
Open Webpage

Weidong Cao, Jiesun Mao, Weixin Zeng, Jianjun Ni

Published: 2025, Last Modified: 16 Jan 2026J. Supercomput. 2025EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: The starfish optimization algorithm (SFOA), a novel meta-heuristic proposed in 2024, is inspired by starfish behaviors like exploration, predation, and regeneration. Despite its good performance in optimization tasks, it still faces issues such as late-stage local optima trapping and premature convergence. To address the issue, a modified starfish optimization algorithm (MSFOA) was designed in this work. First, a brand-new meta-heuristic algorithm improvement strategy based on natural pheromone diffusion, namely, the pheromone attraction strategy, was proposed. Second, the original algorithm’s twisted angle mechanism was improved, and an inertia factor was introduced to modify the out-of-bounds behavior mechanism. Third, a greedy dimension-by-dimension update strategy was adopted to optimize the position update. Finally, a reverse learning phase was introduced, in which reverse learning was undergone by the worst few individuals, while this strategy was also adopted by the remaining individuals with a certain probability. The effectiveness of the algorithm was validated using 33 test functions. Additionally, statistical methods such as the Wilcoxon rank sum test with ties treatment, Cliff’s Delta effect size analysis, 95% confidence interval analysis, multiple test correction, and Friedman statistical test were used to compare and analyze the algorithm with six well-known meta-heuristic algorithms. The results showed that MSFOA ranked first in both test function sets, demonstrating outstanding optimization capabilities. MSFOA also exhibited excellent performance in solving three constrained engineering optimization problems, which further confirms its remarkable effectiveness in practical applications. However, when using MSFOA for a deep neural network tuning, the running time on a regular computer is long, so we introduce the high-performance computing platform to solve this problem and maximize the effectiveness of this algorithm. In summary, MSFOA effectively enhances the performance of the original algorithm and possesses excellent practical application value.