A Probabilistic Niching Evolutionary Computation Framework Based on Binary Space Partitioning.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Evolutionary computation

Background:

• Multimodal optimization problems present challenges in identifying multiple satisfactory solutions.
• Existing search methods often rely on current population data, leading to inefficiencies.
• A need exists for advanced strategies that leverage historical information for better solution discovery.

Purpose of the Study:

• To propose a probabilistic niching evolutionary computation framework for locating diverse and high-quality solutions.
• To enhance search guidance by utilizing comprehensive historical information.
• To provide a universal framework adaptable to various baseline niching algorithms.

Main Methods:

• A binary space partition tree was constructed to organize space-visiting information.
• A probabilistic niching strategy was defined to balance exploration and exploitation using structural historical data.
• The framework was integrated with distance-based differential evolution and topology-based particle swarm optimization algorithms.

Main Results:

• The integrated algorithms demonstrated competitive performance on 20 multimodal optimization test functions.
• The proposed framework significantly improved the ability to find diverse and high-quality solutions.
• The enhanced algorithms outperformed several state-of-the-art niching algorithms.

Conclusions:

• The probabilistic niching evolutionary computation framework effectively guides future searches using historical data.
• This approach enhances the discovery of multiple, high-quality solutions in multimodal optimization.
• The framework offers a versatile and effective enhancement for existing niching algorithms.