Neural Net-Enhanced Competitive Swarm Optimizer for Large-Scale Multiobjective Optimization.

Area of Science:

• Computational Intelligence
• Optimization Algorithms
• Machine Learning

Background:

• Competitive Swarm Optimizer (CSO) shows promise for large-scale multiobjective optimization problems (LMOPs).
• Existing CSO studies often overlook the crucial role of winner particle evolution in final performance.
• A gap exists in enhancing the evolutionary dynamics of winner particles within CSO frameworks.

Purpose of the Study:

• To propose a novel neural network-enhanced CSO (NN-CSO) for improved performance on LMOPs.
• To address the limitation of ignoring winner particle evolution in traditional CSO.
• To leverage neural networks for evolving winner particles and enhancing optimization dynamics.

Main Methods:

• Classifying swarm particles into winner and loser groups via pairwise competition.
• Training a neural network (NN) model using loser particles as input and winner particles as output.
• Evolving winner particles using the trained NN model while loser particles are guided by winners.

Main Results:

• The NN-CSO significantly improves the performance of CSOs on LMOPs.
• Experimental results demonstrate advantages over state-of-the-art large-scale multiobjective evolutionary algorithms.
• The NN model effectively learns and applies promising evolutionary dynamics to winner particles.

Conclusions:

• The proposed NN-CSO effectively enhances winner particle evolution, leading to superior optimization performance.
• NN-CSO offers a viable and improved approach for tackling complex large-scale multiobjective optimization problems.
• This work highlights the potential of integrating neural networks into swarm intelligence for advanced optimization.