The crowding approach to niching in genetic algorithms

Area of Science:

• Evolutionary Computation
• Artificial Intelligence
• Optimization Algorithms

Background:

• Niching techniques are crucial for maintaining diversity in evolutionary algorithms.
• Crowding methods, a subset of niching, aim to preserve multiple optima.
• Local tournament algorithms encompass various strategies including simulated annealing and restricted tournament selection.

Purpose of the Study:

• To introduce and analyze probabilistic crowding as an effective niching technique within local tournament algorithms.
• To develop a general algorithmic and analytical framework applicable to diverse crowding algorithms.
• To investigate the reliable maintenance of subpopulations and predict their behavior.

Main Methods:

• Development of a generalized algorithmic and analytical framework for crowding algorithms.
• Presentation and analysis of the probabilistic crowding niching algorithm.
• Experimental validation of the probabilistic crowding approach and comparison with deterministic crowding.

Main Results:

• Probabilistic crowding is demonstrated to be fast, simple, and requires no additional parameters beyond classical genetic algorithms.
• Reliable maintenance of subpopulations is achieved and shown to be predictable.
• Novel results for deterministic crowding are presented, alongside insights into combining crowding replacement rules and population sizing.

Conclusions:

• Probabilistic crowding offers a robust and efficient method for niching in evolutionary computation.
• The developed framework provides a valuable tool for analyzing and understanding various crowding techniques.
• Experimental results confirm the effectiveness and predictability of probabilistic crowding for maintaining subpopulations.