A Novel Diversity-Based Replacement Strategy for Evolutionary Algorithms.

Area of Science:

• Computer Science
• Artificial Intelligence
• Optimization

Background:

• Premature convergence is a significant limitation in evolutionary algorithms, hindering their performance.
• Maintaining population diversity is a key strategy to mitigate premature convergence.

Purpose of the Study:

• To present a new replacement strategy for evolutionary algorithms designed to preserve useful diversity.
• To address premature convergence by integrating diversity as an explicit objective.

Main Methods:

• The proposed method transforms single-objective problems into multiobjective ones, incorporating diversity as an objective.
• It dynamically adapts the balance between exploration and exploitation across different optimization stages.
• Diversity is measured by calculating distances to the nearest surviving individuals.

Main Results:

• Analyses using a multimodal function validate the design and operational principles.
• Computational results on a packing problem demonstrate significant improvements over existing methods.
• The new approach favorably compares against numerous state-of-the-art optimization schemes.

Conclusions:

• The novel replacement strategy effectively preserves diversity and enhances evolutionary algorithm performance.
• This approach offers a promising solution for overcoming premature convergence in complex optimization tasks.
• The method shows strong potential for application in challenging problems like packing optimization.