A Constrained Learning-Based Competitive Swarm Optimizer for Large-Scale Multiobjective Optimization.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Swarm Intelligence

Background:

• Competitive Swarm Optimizer (CSO) is a key method for large-scale multiobjective optimization problems (LMOPs).
• Existing CSOs using pairwise random competition (PRC) face limitations: poor winner quality hinders learning, and PRC's stochasticity slows convergence.

Purpose of the Study:

• To address limitations of traditional CSOs in solving LMOPs.
• To propose a novel Constrained Learning-based CSO (CL-CSO) for enhanced performance on LMOPs.

Main Methods:

• CL-CSO utilizes reference vectors to partition the objective space into subregions.
• A constrained learning strategy is employed, enabling loser particles to learn from winners within intra- or neighboring subregions.
• A Gaussian model-assisted strategy enhances the diversity and quality of winner particles.

Main Results:

• CL-CSO significantly improves particle learning effects and overall convergence speed.
• Experimental results demonstrate superior performance of CL-CSO compared to existing algorithms on benchmark LMOPs.
• The algorithm effectively handles problems with 2-3 objectives and 500-5000 decision variables, including real-world instance selection problems.

Conclusions:

• The proposed CL-CSO effectively overcomes the limitations of traditional PRC mechanisms in CSOs.
• CL-CSO offers a more efficient and robust approach for tackling large-scale multiobjective optimization problems.
• The method shows promise for practical applications in complex optimization scenarios.