Multi-objective particle swarm optimization with reverse multi-leaders.

Area of Science:

• Computational Intelligence
• Optimization Algorithms

Background:

• Multi-objective particle swarm optimization (MOPSO) algorithms face challenges in balancing solution convergence and diversity.
• Existing MOPSO methods require further improvements for optimal performance.

Purpose of the Study:

• To propose a novel MOPSO algorithm, RMMOPSO, that improves the balance between convergence and diversity.
• To enhance the learning process and guidance of particles towards the true Pareto front.

Main Methods:

• Developed a reverse multi-leaders multi-objective particle swarm optimization (RMMOPSO) algorithm.
• Introduced a convergence strategy using global ranking and a diversity strategy based on mean angular distance.
• Implemented a reverse selection method for choosing two global leaders and an information fusion strategy for updating personal best.
• Proposed a new particle velocity updating method for better convergence-diversity balance.

Main Results:

• RMMOPSO demonstrated superior comprehensive performance compared to state-of-the-art MOPSOs and multi-objective evolutionary algorithms (MOEAs).
• Simulations on 22 benchmark problems validated the effectiveness of the proposed strategies.

Conclusions:

• RMMOPSO effectively balances convergence and diversity in multi-objective optimization.
• The proposed reverse multi-leaders approach and associated strategies significantly improve algorithm performance.