Particle Swarm Optimization with Double Learning Patterns

Area of Science:

• Computational Intelligence
• Swarm Intelligence
• Optimization Algorithms

Background:

• Particle Swarm Optimization (PSO) is widely used for optimization problems.
• PSO often faces premature convergence due to loss of swarm diversity.

Purpose of the Study:

• To address PSO's premature convergence by enhancing swarm diversity.
• To develop a PSO variant that balances convergence speed and diversity.

Main Methods:

• Analysis of swarm motion based on learning parameter probabilities.
• Development of a PSO with double learning patterns (PSO-DLP) using master and slave swarms.
• Implementation of an interaction mechanism between swarms to improve exploration and exploitation.

Main Results:

• PSO-DLP demonstrated a trade-off between convergence speed and swarm diversity.
• The interaction mechanism helped swarms escape local optima and refine solutions.
• Evaluated on 20 benchmark functions, PSO-DLP outperformed eight other PSO variants.

Conclusions:

• PSO-DLP offers a promising approach to overcome premature convergence in PSO.
• The dual-swarm strategy effectively maintains diversity and improves solution precision.
• PSO-DLP shows superior performance on complex and rotated multimodal optimization problems.