A particle swarm optimization variant with an inner variable learning strategy.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Particle Swarm Optimization (PSO) is effective for global optimization but struggles with high-dimensional and complex problems.
• Existing PSO variants often lack specialized strategies for specific problem structures.
• Integrating domain knowledge can enhance the performance of evolutionary algorithms.

Purpose of the Study:

• To develop a novel PSO variant (PSO-IVL) that incorporates problem-oriented knowledge.
• To improve the efficiency of PSO for functions with symmetric variables.
• To enhance the ability of PSO to escape local optima in complex optimization landscapes.

Main Methods:

• Introduced an Inner Variable Learning (IVL) strategy that exploits quantitative relations among symmetric variables.
• Developed a novel, adaptive trap detection and jumping out strategy for individual particles.
• Tested the proposed PSO-IVL algorithm on representative optimization functions.

Main Results:

• PSO-IVL demonstrated superior efficiency in optimizing functions with symmetric variables.
• The IVL strategy effectively guided particles by identifying and learning from exemplar variables.
• The trap detection and jumping out mechanism successfully helped particles avoid local optima.
• Experimental simulations confirmed the excellent performance of PSO-IVL compared to existing methods.

Conclusions:

• The proposed PSO-IVL algorithm significantly enhances optimization performance, particularly for problems with symmetric variables.
• Augmenting evolutionary algorithms with problem-oriented domain knowledge is a viable and effective approach.
• PSO-IVL offers a promising solution for complex, high-dimensional optimization tasks.