Virtual Position Guided Strategy for Particle Swarm Optimization Algorithms on Multimodal Problems.

Area of Science:

• Computational Intelligence
• Swarm Intelligence
• Optimization Algorithms

Background:

• Premature convergence is a significant challenge in particle swarm optimization (PSO), particularly for multimodal problems requiring sustained swarm diversity.
• Existing diversity-guided strategies often fail to fully resolve the issue of premature convergence in PSO.
• Maintaining swarm diversity is critical for effective exploration and exploitation in PSO algorithms.

Purpose of the Study:

• To propose a novel Virtual Position Guided (VPG) strategy to address premature convergence in particle swarm optimization (PSO).
• To enhance the performance of PSO algorithms on multimodal optimization problems by improving swarm diversity.
• To develop a general and adaptable strategy applicable to various PSO algorithms.

Main Methods:

• The Virtual Position Guided (VPG) strategy calculates diversity values for two populations and establishes a diversity baseline.
• It dynamically guides the algorithm through three phases (divergence, normal, acceleration) based on diversity values and the baseline.
• The strategy balances exploration and exploitation to steer the algorithm away from local optima.

Main Results:

• Experimental results show the VPG strategy outperforms canonical and state-of-the-art diversity guidance strategies.
• The VPG strategy effectively improves the search performance of most PSO algorithms on multimodal problems.
• The strategy demonstrates effectiveness across various dimensionalities of multimodal problems.

Conclusions:

• The proposed Virtual Position Guided (VPG) strategy is a superior method for enhancing particle swarm optimization (PSO) performance.
• VPG effectively mitigates premature convergence by dynamically managing swarm diversity and balancing exploration-exploitation.
• The strategy's adaptability and ease of implementation make it a valuable contribution to swarm intelligence research.