Autonomous Parameter Balance in Population-Based Approaches: A Self-Adaptive Learning-Based Strategy.

Area of Science:

• Computational Intelligence
• Operations Research
• Computer Science

Background:

• Population-based metaheuristics are widely used for optimization but struggle with parameter control, particularly population size.
• Balancing solution quality and computational time is a persistent challenge, especially for novel optimization problems.

Purpose of the Study:

• To propose a novel self-adaptive strategy for dynamically adjusting population size in population-based metaheuristics.
• To enhance the performance and search process of these algorithms through on-line population balancing.

Main Methods:

• A three-component approach: optimization-based, learning-based, and probabilistic-based selector.
• The strategy dynamically adjusts population size based on real-time data and learning.
• Extensive experiments were conducted on Manufacturing Cell Design, Set Covering, and Multidimensional Knapsack problems.

Main Results:

• The proposed self-adaptive strategy demonstrates competitive performance against established methods.
• It effectively balances solution quality and computational efficiency.
• The approach shows promise for improving the search process in discrete optimization.

Conclusions:

• The self-adaptive strategy offers an effective method for dynamic population size adjustment in metaheuristics.
• It provides a robust solution for optimizing complex discrete problems.
• Future work may explore dynamic adjustment of interacting solution numbers within the search process.