Multimodal Optimization Enhanced Cooperative Coevolution for Large-Scale Optimization.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Evolutionary computation

Background:

• Cooperative coevolutionary (CC) algorithms divide problems into subcomponents for large-scale optimization.
• Inaccurate decomposition can lead to information loss and hinder performance.
• Existing CC methods struggle with complex fitness landscape topologies.

Purpose of the Study:

• To propose a novel CC algorithm that addresses the limitations of inaccurate decomposition.
• To improve the handling of interacting decision variables in large-scale optimization.
• To enhance the search for multiple optima within subcomponents.

Main Methods:

• Incorporating a multimodal optimization procedure into each subcomponent, adaptively triggered by optimizer status.
• Utilizing informative representatives (multiple optima) exchanged among subcomponents.
• Employing a nondominance-based selection scheme to construct complete solutions.

Main Results:

• Demonstrated superior performance against five popular CC algorithms on hard problems.
• Validated enhanced performance through a comprehensive comparison with 17 state-of-the-art CC and metaheuristic algorithms.
• Showcased effectiveness on 20 1000-dimensional benchmark functions.

Conclusions:

• The proposed CC algorithm effectively compensates for decomposition inaccuracies.
• The method offers a significant advancement in large-scale optimization using cooperative coevolution.
• The approach provides a robust strategy for complex optimization landscapes.