Self-organizing map based differential evolution with dynamic selection strategy for multimodal optimization...

Area of Science:

• Computational intelligence
• Optimization algorithms
• Evolutionary computation

Background:

• Multimodal optimization problems (MMOPs) present challenges in balancing population diversity and convergence.
• Effective niching techniques are crucial for locating multiple global optima in MMOPs.
• Existing differential evolution (DE) algorithms may struggle with the complexities of MMOPs.

Purpose of the Study:

• To propose a novel algorithm, SOMDE-DS, to improve the performance of DE in solving MMOPs.
• To enhance the ability of DE to balance population diversity and convergence.
• To effectively locate multiple global optima and refine their accuracy.

Main Methods:

• A Self-Organizing Map (SOM) is employed as a niching technique for reasonable population division based on individual similarity.
• A Variable Neighborhood Search (VNS) strategy is integrated to expand the search space and identify more potential optimal regions.
• A Dynamic Selection (DS) strategy is designed to balance exploration and exploitation by combining local and global search approaches.

Main Results:

• The proposed SOMDE-DS algorithm was evaluated on benchmark CEC'2013 datasets.
• Experimental results demonstrate that SOMDE-DS outperforms or is competitive with existing widely used multimodal optimization algorithms.
• The method shows significant improvements in locating global optima and refining solution accuracy for MMOPs.

Conclusions:

• SOMDE-DS offers a robust and effective approach for addressing multimodal optimization problems.
• The integration of SOM, VNS, and DS strategies enhances the performance of differential evolution.
• The proposed algorithm provides a valuable contribution to the field of evolutionary computation for complex optimization tasks.