Robust Multiobjective Particle Swarm Optimization With Feedback Compensation Strategy.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Evolutionary Computation

Background:

• Multiobjective particle swarm optimization (MOPSO) is effective for multiobjective problems (MOPs).
• Random selection of parameters and leaders in MOPSO introduces uncertainty, degrading performance.
• Addressing evolutionary process uncertainty is crucial for improving MOPSO.

Purpose of the Study:

• To propose a robust MOPSO with feedback compensation (RMOPSO-FC).
• To introduce a closed-loop optimization framework to mitigate uncertainty's negative impact.
• To enhance the search capability and algorithmic robustness of MOPSO.

Main Methods:

• Establishing Gaussian process (GP) models using dynamically updated archives for particle posterior distribution.
• Collecting feedback information on particle evolution.
• Designing an intergenerational binary metric to evaluate evolutionary potential and identify negative evolutionary directions.
• Implementing a compensation mechanism to correct negative particle evolution by modifying the update paradigm.

Main Results:

• The proposed RMOPSO-FC effectively reduces the negative influence of uncertainty.
• Feedback compensation guides particles toward the true Pareto front (PF).
• Comparative simulations demonstrate superior search capability for PFs.
• RMOPSO-FC exhibits enhanced algorithmic robustness over multiple runs.

Conclusions:

• RMOPSO-FC offers a novel approach to address uncertainty in MOPSO.
• The feedback compensation mechanism improves the exploration of the Pareto front.
• The method provides superior performance and robustness compared to existing approaches.