Pareto-like sequential sampling heuristic for global optimisation.

Area of Science:

• Computational intelligence
• Optimization algorithms
• Heuristics

Background:

• Traditional metaheuristics often rely on mutation or crossover operations.
• Premature convergence is a common challenge in global optimization.
• Balancing exploration and exploitation is crucial for effective optimization.

Purpose of the Study:

• To propose a novel global optimization algorithm.
• To introduce a self-adaptive mechanism for dynamic domain tightening.
• To avoid premature convergence and minimize structural bias.

Main Methods:

• Sequential random sampling for diversification and intensification.
• Self-adaptive mechanism controlling prominent search domains.
• Theoretical analysis of exploration bias and diversification rate.

Main Results:

• The algorithm performs well on standard and engineering optimization problems.
• It excels in finding global minima for high-dimensional, non-convex, and multimodal functions.
• Outperforms recent algorithms on CEC2017 composite problems, especially under limited iterations.

Conclusions:

• The proposed algorithm offers an unbiased exploration with a constant diversification rate.
• It effectively balances diversification and intensification.
• Its simple design facilitates hybridization with other search paradigms.